LibreSocial · Documentation

Publication Archive

All documents of the LibreSocial / LifeSocial.KOM project archive (papers, theses, project reports, slides, manuals), 2001–2025. Sort by year or title, filter by architecture area, expand abstracts, select documents and download their PDFs or a combined BibTeX file. The table below is plain HTML — filtering and downloads are progressive enhancement.

Year Title TypeArchitecture areasLinks
2001 PAST: A large-scale, persistent peer-to-peer storage utility
Peter Druschel (Rice University, Houston TX), Antony Rowstron (Microsoft Research, Cambridge UK)
abstract

This position paper sketches the design of PAST, a large-scale, Internet-based global storage utility that is entirely self-organizing. PAST nodes serve as access points for clients, route client insert and retrieve requests, and contribute storage; inserted files are replicated on several nodes so that, with high probability, the replica set is diverse with respect to geographic location, ownership, administration, network connectivity and rule of law. Three aspects of the design are emphasized: the Pastry location and routing scheme, which reliably routes client requests with good network locality and automatically resolves node additions and failures; the use of randomization to ensure diversity among the nodes storing a file's replicas and to provide load balancing; and the optional use of smartcards held by users and issued by a broker, which support a quota system balancing the supply and demand of storage. Because a file is bound to a quasi-unique fileId created at insertion time, stored files are immutable and there is no delete operation but only reclamation of the storage, which avoids agreement protocols among the nodes storing the file. The paper also positions PAST against related work such as Napster, Gnutella, Freenet, OceanStore, FarSite, FreeHaven, Eternity, Chord, Tapestry and CAN.

Publikation (Konferenzpapier, HotOS VIII, Schloss Elmau, Mai 2001) Overlay (Pastry) Storage & DSS Monitoring Security Plugin system App & Kernel PDF · BibTeX
2001 Pastry: Scalable, decentralized object location and routing for large-scale peer-to-peer systems
Antony Rowstron (Microsoft Research Ltd, Cambridge UK), Peter Druschel (Rice University, Houston TX)
abstract

This paper presents the design and evaluation of Pastry, a scalable, distributed object location and routing substrate for wide-area peer-to-peer ap- plications. Pastry performs application-level routing and object location in a po- tentially very large overlay network of nodes connected via the Internet. It can be used to support a variety of peer-to-peer applications, including global data storage, data sharing, group communication and naming. Each node in the Pastry network has a unique identifier (nodeId). When presented with a message and a key, a Pastry node efficiently routes the message to the node with a nodeId that is numerically closest to the key, among all currently live Pastry nodes. Each Pastry node keeps track of its immediate neighbors in the nodeId space, and notifies applications of new node arrivals, node failures and recoveries. Pastry takes into account network locality; it seeks to minimize the distance messages travel, according to a to scalar proximity metric like the number of IP routing hops. Pastry is completely decentralized, scalable, and self-organizing; it automatically adapts to the arrival, departure and failure of nodes. Experimental results obtained with a prototype implementation on an emulated network of up to 100,000 nodes confirm Pastry’s scalability and efficiency, its ability to self-organize and adapt to node failures, and its good network locality properties.

Publikation (Proc. 18th IFIP/ACM International Conference on Distributed Systems Platforms, Middleware 2001, Heidelberg) Overlay (Pastry) Storage & DSS Messaging Monitoring NAT traversal Testing & Evaluation PDF · BibTeX
2001 Storage Management and Caching in PAST, a Large-Scale, Persistent Peer-to-Peer Storage Utility
Antony Rowstron (Microsoft Research, Cambridge), Peter Druschel (Rice University)
abstract

This paper presents and evaluates the storage management and caching of PAST, a large-scale peer-to-peer persistent storage utility built on a self-organizing Internet-based overlay network of storage nodes that cooperatively route file queries, store multiple replicas and cache additional copies of popular files. Storage nodes and files carry uniformly distributed identifiers and replicas are stored at the nodes whose identifiers match a file's identifier most closely, which only approximately balances load because node capacities and file sizes are non-uniform. PAST therefore balances storage explicitly through replica diversion to leaf-set neighbours and file diversion with a new salt, using an acceptance policy based on file size relative to free space with separate thresholds for primary and diverted replicas: without diversion 51.1 per cent of insert attempts fail and global utilization reaches only 60.8 per cent, while with the chosen thresholds utilization rises to 94.9 per cent (leaf set size 16) and 98.2 per cent (size 32) at failure rates of 2.4 and 0.7 per cent. Caching of popular files uses a GreedyDual-Size insertion and replacement policy that outperforms LRU in hit rate and fetch distance. Extensive trace-driven experiments show that the system minimizes fetch distance, balances the query load for popular files and degrades gracefully as global storage utilization increases beyond 95 per cent.

Publikation (ACM SOSP 2001) Overlay (Pastry) Storage & DSS PDF · BibTeX
2008 A Distributed Platform for Multimedia Communities
Kalman Graffi, Sergey Podrajanski, Patrick Mukherjee, Aleksandra Kovacevic, Ralf Steinmetz (TU Darmstadt, KOM und RTS; DFG-Forschungsgruppe 733 „QuaP2P“)
abstract

Online community platforms and multimedia content delivery are merging, and current platforms such as Facebook and YouTube are client-server based, which results in high administration costs for the provider, while peer-to-peer systems offer scalability and low cost but are limited in functionality. The paper identifies the key challenges of this new application of the peer-to-peer paradigm and presents a plugin-based, easily extendible and multifunctional framework for multimedia online communities, implemented in Java on FreePastry as the distributed hash table and PAST as the storage and replication layer. It shows how the data are organized with distributed linked lists for user profiles, friend lists, groups, photo albums and photos, where keys are derived by concatenation and hashing and every object carries metadata plus pointers to its successor objects, thus spreading load over many peers. The layered framework consists of the overlay, storage and replication, a storage dispatcher and a message dispatcher, an information cache that represents the distributed objects locally in available, pending or missing states with timestamps and LRU replacement so the GUI can always show a consistent snapshot of incomplete results, the plug-in layer and the user interface. Design decisions cover pseudonymity with node identifiers based on elliptic-curve public keys, author-signed objects and symmetric encryption for confidential data, and request-based rather than event-triggered event propagation that keeps plug-ins simple and stateless.

Publikation (IEEE-Konferenzpaper, ISM 2008) Storage & DSS Messaging Monitoring Search Plugin system App & Kernel PDF · BibTeX
2008 A Peer-to-Peer based Framework for Large-Scale Social Networking Platforms
Sergey Podrajanski (Betreuer: Prof. Dr.-Ing. Ralf Steinmetz, Tutor: Kálmán György Graffi; KOM-D-0316, TU Darmstadt, Multimedia Communications Lab)
abstract

CommandHandler is an abstract class which implements the CommandHan- dler interface and all methods except the method ’executeCommand(arguments)’. Every command class extends this abstract class and implements the only abstract method - ’executeCommand(arguments)’. 5.4. USER INTERFACE 105 Commands class offers method for registration of single command class implementa- tions. All concrete commands implementations are registered in the Commands class and can be used by the user. The described framework for creating a text-based user interface is easily extendable. New commands classes have to provide a name of the new command and the action which is processed when the command is executed. 106 CHAPTER 5. PROTOTYPE DESIGN AND IMPLEMENTATION Chapter 6 Architecture Evaluation In this section, an evaluation of the Peer-to-Peer based architecture presented in this diploma thesis is given. First, the developed Peer-to-Peer based architecture is compared with a typical client/server based architecture concerning robustness, storage behavior, scalability, administration costs and security issues. Then the developed Peer-to-Peer ar- chitecture will be evaluated concerning extendability and event propagation.

Studienarbeit (Diplomarbeit, "Diploma Thesis", eingereicht 15. Juni 2008) Overlay (Pastry) Storage & DSS Messaging Search Plugin system App & Kernel PDF · BibTeX
2008 Design and Implementation of a Service-oriented, Dynamic and Modular P2P-based Multimedia Online Community Platform using OSGi
Luciana Mara de Carvalho Alvite; Betreuer: Prof. Dr.-Ing. Ralf Steinmetz; Tutor: Kálmán György Graffi (KOM, TU Darmstadt; Prüfungsnummer KOM-S-0279)
abstract

This Studienarbeit analyses the rigid structure of the LifeSocial.KOM platform, which was delivered as a single JAR in which every plug-in was registered statically and loaded at start-up, so that any change required recompilation and a restart, and in which JARs offer neither versioned dependencies, runtime dependency resolution nor information hiding. The platform is rebuilt as a service-oriented, dynamic and modular system on OSGi Release 4, with Eclipse Equinox chosen after comparing alternatives such as Knopflerfish and commercial runtimes. The work describes the resulting bundle structure - logger, core, utilities, network, file system, local storage, storage with information cache and storage dispatcher, message dispatcher, e-mail sender, user interface and the plug-ins - together with bundle activation, the service registry and declarative services with cardinalities and static or dynamic policies. A command architecture is introduced in which commands become active together with their plug-in and return status strings for a later graphical user interface, an extension mechanism is demonstrated with a blog plug-in, and a cyclic dependency between the login and profile plug-ins is removed by moving createUserProfile to the command level. The appendices list the console commands and describe how to start the platform in test and live mode; future work mentioned includes a plug-in repository, a graphical user interface and missing functions such as blogs and videos.

Studienarbeit Overlay (Pastry) Storage & DSS Messaging Search Plugin system App & Kernel PDF · BibTeX
2008 LifeSocial – Masterpraktikum: Entwurf und Implementierung neuer Plug-ins
Florian Gattung, Martin Kleine, Immanuel Pahlke (betreut von Kalman Graffi, Multimedia Communications Lab, TU Darmstadt)
abstract

This report documents the master lab exercise on LifeSocial in the winter term 2008/2009 organized by the Multimedia Communications Lab at TU Darmstadt, in which a three-person team analysed the platform and designed and implemented new plug-ins intended for the presentation at CeBIT 2009. The approach section describes the kick-off meeting on 15 October 2008 and the subsequent milestone meetings, and the analysis of the existing plug-ins and of candidates for new ones, keeping business and private clients in view. Because most candidates turned out to share group functionality, the team factored it out into the abstract base class GroupEnabledPlugin, which provides joinGroup and leaveGroup, message distribution and persisted group storage, with data isolated per plug-in instance and group state kept in the DHT as shared items for group names, group members and message history. Eight plug-ins were implemented: the group base, group chat, whiteboard, TicTacToe, voting, a schedule plug-in preparing RFC-2445/iCal support, an address book with vCard import and export, and a file transfer plug-in that stores shared file path lists per user and transfers file contents as byte arrays inside messages sent through the message dispatcher, caching undeliverable messages in the DHT. The report also covers additional activities: fixing a messaging fault introduced by the new security extension, a createenvironment command that automatically builds a test environment with users, groups and friendships, and UML documentation plus a screencast for the Eclipse and SVN setup.

Projektbericht (Praktikums-/Docu-Writing-Bericht) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2008 SkyEye.KOM: An Information Management Over-Overlay for Getting the Oracle View on Structured P2P Systems
Kalman Graffi, Aleksandra Kovacevic, Song Xiao, Ralf Steinmetz (TU Darmstadt, Multimedia Communications Lab KOM)
abstract

To ease the development and maintenance of complex P2P applications that combine several P2P functionalities such as streaming and dependable storage, this paper suggests extending structured P2P systems with a dedicated information management layer. SkyEye.KOM is such an information management over-overlay applicable as an additional layer on any distributed hash table: it generates statistics on the whole P2P network and provides capacity-based peer search, so that the individual functionality layers of an application can find suitable peers for layer-specific role assignment. The identifier space is recursively divided into domains whose coordinators are determined deterministically from a domain key and the responsibility information of the underlying DHT, which yields a tree of O(log N) depth that needs virtually no maintenance of its own. Updates are pushed bottom-up and carry an aggregatable statistics part with averages, standard deviations and confidence intervals plus a non-aggregatable capacity part covering CPU, bandwidth, storage and online time with a time-to-live, while queries for n peers with AND, OR, less-than or greater-than constraints over these attributes are answered as they travel up the tree; load balancing uses thresholds and support peers to relieve weak nodes. Evaluation by simulation in PeerfactSim.KOM with 1000, 5000 and 10000 peers shows query answers after zero to four hops on average, about 98.5 per cent of the peers in result lists online, one to three update messages per peer per update interval, and overall very low maintenance cost.

Publikation (IEEE ICPADS 2008; der Dateiname nennt eine ältere Titelfassung „Towards an Efficient Management Architecture") Overlay (Pastry) Storage & DSS Monitoring PDF · BibTeX
2009 Development and Evaluation of a Framework for Monitoring Peer-to-Peer Systems
Christian Groß (Betreuer: Prof. Dr.-Ing. Ralf Steinmetz; Tutor: Kálmán György Graffi)
abstract

This master thesis develops and evaluates a framework for distributed information management systems that makes structured peer-to-peer systems monitorable and searchable for peer capacities. It begins with a survey of state-of-the-art systems such as Astrolabe, Willow, SOMO, SDIMS, CONE, T-Man and proactive epidemic aggregation and derives a taxonomy according to communication mechanism and aggregation topology, complemented by properties such as overlay independence, churn resistance, load balancing and security, noting that most published approaches neglect security. The prototype implements SkyEye.KOM, a single-tree approach that gathers system statistics and answers capacity-based queries such as for ten peers with more than 10 MB of free disk space; it is overlay independent because mapping functions map the entire identifier space of the underlying DHT onto its own identifier space in the interval from 0 to 1, and it balances load by merging children below a threshold and by support peers for weak nodes. For extensibility and easy integration the framework was built as OSGi bundles that communicate only through interfaces, and it was integrated into the two OSGi-based applications LifeSocial.KOM and the distributed Eclipse IDE PIPE. The evaluation in a testbed of up to 500 peers shows resistance to churn of up to 50 per cent, an average of about 3 KB/s per peer and about two hops with a five-second update interval, up to 70 per cent traffic reduction through compression, and roughly 200 bytes per second for intervals of 30 to 60 seconds, while also revealing that concatenation-based capacity aggregation makes bandwidth grow towards the upper tree levels, so that the approach does not scale without filter mechanisms or incremental updates.

Masterarbeit (KOM-D-351, TU Darmstadt, Multimedia Communications Lab) Overlay (Pastry) Storage & DSS Messaging Monitoring App & Kernel PDF · BibTeX
2009 Development of LifeSocial Plug-ins (Quick Start)
Luciana Alvite (Betreuer: Prof. Dr.-Ing. Ralf Steinmetz, Tutor: Kalman Graffi)
abstract

Commands (see de.tud.kom.lifesocial.core.commands) already have them, so nothing needs to be added to MyPluginCommands. Listings 3.11 and 3.12 shows the binding methods for MyPluginComponent and Abstract- MyPluginCommandHandler respectively

Studienarbeit / HowTo-Report (KOM – Multimedia Communications Lab, TU Darmstadt) Storage & DSS Messaging Plugin system App & Kernel PDF · BibTeX
2009 LifeSocial Commands – Report (Kommando-Referenz der LifeSocial.KOM-Plattform)
Luciana (Mara de Carvalho) Alvite; Betreuer: Prof. Dr.-Ing. Ralf Steinmetz, Tutor: Kálmán Graffi (KOM, TU Darmstadt)
abstract

This report is the command reference of the console interface of the OSGi-based LifeSocial.KOM platform, listing all commands available on 28 September 2009 plug-in by plug-in: network, login, profile, friends, groups, live chat, messaging, search, photos and whiteboard. Each command table states the arguments, a description and an example; the network commands are Connect, ConnectP, Disconnect and IsConnected, the default port 9001 being defined in the PropertiesContainer class of the utilities module, and the login commands presuppose an established network connection, with registration logging the user in automatically. The asynchronous read semantics of the information cache become visible through the loaded-queries such as ProfileCompletelyLoaded, FriendsListLoaded and AlbumsLoaded. One appendix documents the ten string profile fields with username, email, city and country as mandatory entries and notes that EditProfileField only stages a change while StoreMyProfile writes it to the network, with no plausibility check of the input; another appendix lists the status messages of the network and login plug-ins. A second part explains how to start the platform: in test mode Connect without arguments builds a new ring on localhost and the default folder containing the default image must exist, whereas live operation over the Internet requires binding to the address of an existing ring node and setting or deleting the FreePastry parameter nat_search_policy.

Studienarbeit / HowTo-Dokumentation (Report) Overlay (Pastry) Storage & DSS Messaging Search Plugin system App & Kernel PDF · BibTeX
2009 Monitoring and Management of P2P Overlays — How to coordinate millions of autonomous peers to provide controlled quality of service?
Kalman Graffi (Dipl.-Math. Dipl.-Inform., TU Darmstadt, Multimedia Communications Lab KOM, Prof. Dr.-Ing. Ralf Steinmetz)
abstract

This IEEE P2P 2009 conference paper, its poster and the accompanying conference talk present an approach to the monitoring and management of structured P2P overlays, motivated by the question of how millions of autonomous peers can be coordinated to provide a controlled quality of service. Quality goals are given as intervals, for instance a hop count between 7 and 10 or a maximum standard deviation of 500 per cent for peer load, so the overlay must reconfigure itself automatically in the face of churn. SkyEye.KOM is introduced as a proactive monitoring and management layer that forms a tree-based over-overlay instead of being integrated into the overlay, uses equal peer roles and deterministic identifiers that are resolved dynamically through the DHT, and aggregates periodic updates at every node into mean, standard deviation, variance, sum, minimum and maximum so that message size stays independent of the number of samples. The root of the tree derives a new configuration from the aggregated global view and distributes it downwards to all peers, with protection against oscillation by making changes effective, delaying execution and acting only when the slope of the value history is small. Simulations in PeerfactSim.KOM with an ideal DHT, 5000 nodes, a network-delay model and a churn model based on KAD measurements assess relative monitoring error, information age and traffic overhead, and a Chord case study varying the finger-table size shows it being doubled from 20 to 80 or reduced from 160 to 116 until the hop-count target is reached and maintained; the outlook mentions applying the cycle to Kademlia and learning configuration rules automatically with machine learning or genetic algorithms.

Vortrag (Conference-Talk/Folien, IEEE P2P 2009, 20. August 2009) Overlay (Pastry) Storage & DSS Messaging Monitoring App & Kernel PDF · BibTeX
2009 Monitoring and Management of Structured Peer-to-Peer Overlays
Kalman Graffi, Dominik Stingl, Julius Rückert, Aleksandra Kovacevic, Ralf Steinmetz (TU Darmstadt, Multimedia Communications Lab KOM)
abstract

This IEEE P2P 2009 conference paper, its poster and the accompanying conference talk present an approach to the monitoring and management of structured P2P overlays, motivated by the question of how millions of autonomous peers can be coordinated to provide a controlled quality of service. Quality goals are given as intervals, for instance a hop count between 7 and 10 or a maximum standard deviation of 500 per cent for peer load, so the overlay must reconfigure itself automatically in the face of churn. SkyEye.KOM is introduced as a proactive monitoring and management layer that forms a tree-based over-overlay instead of being integrated into the overlay, uses equal peer roles and deterministic identifiers that are resolved dynamically through the DHT, and aggregates periodic updates at every node into mean, standard deviation, variance, sum, minimum and maximum so that message size stays independent of the number of samples. The root of the tree derives a new configuration from the aggregated global view and distributes it downwards to all peers, with protection against oscillation by making changes effective, delaying execution and acting only when the slope of the value history is small. Simulations in PeerfactSim.KOM with an ideal DHT, 5000 nodes, a network-delay model and a churn model based on KAD measurements assess relative monitoring error, information age and traffic overhead, and a Chord case study varying the finger-table size shows it being doubled from 20 to 80 or reduced from 160 to 116 until the hop-count target is reached and maintained; the outlook mentions applying the cycle to Kademlia and learning configuration rules automatically with machine learning or genetic algorithms.

Poster (zu IEEE P2P 2009) Overlay (Pastry) Storage & DSS Messaging Monitoring PDF · BibTeX
2009 Monitoring and Management of Structured Peer-to-Peer Systems
Kalman Graffi, Dominik Stingl, Julius Rückert, Aleksandra Kovacevic, Ralf Steinmetz (TU Darmstadt, Multimedia Communications Lab KOM)
abstract

This IEEE P2P 2009 conference paper, its poster and the accompanying conference talk present an approach to the monitoring and management of structured P2P overlays, motivated by the question of how millions of autonomous peers can be coordinated to provide a controlled quality of service. Quality goals are given as intervals, for instance a hop count between 7 and 10 or a maximum standard deviation of 500 per cent for peer load, so the overlay must reconfigure itself automatically in the face of churn. SkyEye.KOM is introduced as a proactive monitoring and management layer that forms a tree-based over-overlay instead of being integrated into the overlay, uses equal peer roles and deterministic identifiers that are resolved dynamically through the DHT, and aggregates periodic updates at every node into mean, standard deviation, variance, sum, minimum and maximum so that message size stays independent of the number of samples. The root of the tree derives a new configuration from the aggregated global view and distributes it downwards to all peers, with protection against oscillation by making changes effective, delaying execution and acting only when the slope of the value history is small. Simulations in PeerfactSim.KOM with an ideal DHT, 5000 nodes, a network-delay model and a churn model based on KAD measurements assess relative monitoring error, information age and traffic overhead, and a Chord case study varying the finger-table size shows it being doubled from 20 to 80 or reduced from 160 to 116 until the hop-count target is reached and maintained; the outlook mentions applying the cycle to Kademlia and learning configuration rules automatically with machine learning or genetic algorithms.

Publikation (IEEE P2P 2009) Overlay (Pastry) Storage & DSS Monitoring App & Kernel PDF · BibTeX
2009 Practical Security in P2P-based Social Networks
Kalman Graffi, Patrick Mukherjee, Burkhard Menges, Daniel Hartung, Aleksandra Kovacevic, Ralf Steinmetz (TU Darmstadt: Multimedia Communications Lab, Real-Time Systems Lab; Norwegian Information Security Laboratory / Gjøvik University College)
abstract

This IEEE LCN 2009 paper and its conference poster present a practical, lightweight security framework for DHT-based social network platforms such as LifeSocial.KOM, which must replace the trust and administration functions that central servers provide in client/server networks. The solution establishes a public key infrastructure without servers or certificate authorities: registration derives an asymmetric key pair from the chosen user name and passphrase, the public key becomes the nodeID and userID of the user, and a signed LoginItem announces the user's presence so that any peer can verify the identity and encrypt traffic towards that user. Access control is ACL based rather than capability based, because ACLs can be attached to the data object itself and survive replication and caching: every sensitive SharedItem is encrypted with an object-specific symmetric key, and copies of that key wrapped with the public keys of the privileged readers are attached as a key list, forming a signed CryptedItem that any node may store, replicate and cache but only privileged users can decrypt; group access works the same way with a single group key held in a key list signed by the group administrator. Messaging and live chat use a hybrid scheme in which each session establishes a symmetric key wrapped with the receiver's public key, and every message is signed for integrity and sender authentication. A testbed evaluation (FreePastry with the identifier space enlarged to 1088 bits, RSA with 1024-bit keys, AES with 128-bit keys, averages over 100 runs on a Core 2 Quad) reports a nearly constant absolute message overhead below 2 KB and roughly 10 ms for encryption and decryption, while the overhead of an item grows with the number of privileged users, about 413 bytes per additional reader and 89 ms encryption time for 200 of them, which is why group-based access control is recommended beyond that size. The poster summarizes the same contributions and adds the outlook of quality monitoring and management, deletion enforcement, detection of illegal content, further interaction plugins and an alpha/beta deployment.

Poster (Konferenzposter zu IEEE LCN 2009) Overlay (Pastry) Storage & DSS Messaging Security Search Plugin system App & Kernel PDF · BibTeX
2009 Practical Security in P2P-based Social Networks (IEEE LCN 2009)
Kalman Graffi, Patrick Mukherjee, Burkhard Menges, Daniel Hartung, Aleksandra Kovacevic, Ralf Steinmetz (Multimedia Communications Lab und Real-Time Systems Lab, TU Darmstadt; Norwegian Information Security Laboratory, Gjøvik University College); gefördert durch DFG-Forschergruppe 733 „QuaP2P"
abstract

This IEEE LCN 2009 paper and its conference poster present a practical, lightweight security framework for DHT-based social network platforms such as LifeSocial.KOM, which must replace the trust and administration functions that central servers provide in client/server networks. The solution establishes a public key infrastructure without servers or certificate authorities: registration derives an asymmetric key pair from the chosen user name and passphrase, the public key becomes the nodeID and userID of the user, and a signed LoginItem announces the user's presence so that any peer can verify the identity and encrypt traffic towards that user. Access control is ACL based rather than capability based, because ACLs can be attached to the data object itself and survive replication and caching: every sensitive SharedItem is encrypted with an object-specific symmetric key, and copies of that key wrapped with the public keys of the privileged readers are attached as a key list, forming a signed CryptedItem that any node may store, replicate and cache but only privileged users can decrypt; group access works the same way with a single group key held in a key list signed by the group administrator. Messaging and live chat use a hybrid scheme in which each session establishes a symmetric key wrapped with the receiver's public key, and every message is signed for integrity and sender authentication. A testbed evaluation (FreePastry with the identifier space enlarged to 1088 bits, RSA with 1024-bit keys, AES with 128-bit keys, averages over 100 runs on a Core 2 Quad) reports a nearly constant absolute message overhead below 2 KB and roughly 10 ms for encryption and decryption, while the overhead of an item grows with the number of privileged users, about 413 bytes per additional reader and 89 ms encryption time for 200 of them, which is why group-based access control is recommended beyond that size. The poster summarizes the same contributions and adds the outlook of quality monitoring and management, deletion enforcement, detection of illegal content, further interaction plugins and an alpha/beta deployment.

Publikation (IEEE LCN 2009, als „RELEVANT" markierter Volltext) Overlay (Pastry) Storage & DSS Messaging Monitoring Security Search Plugin system App & Kernel PDF · BibTeX
2009 Updating the core LifeSocial library (How-To)
Luciana Alvite (Betreuer: Prof. Dr.-Ing. Ralf Steinmetz, Tutor: Kalman Graffi)
abstract

This short companion document to a Studienarbeit at the Multimedia Communications Lab (KOM) of TU Darmstadt is a step-by-step Eclipse how-to for exporting the non-OSGi core LifeSocial library as a runnable JAR file. It explains how to right-click the project "LifeSocial core", choose Export and then Jar File, tick the options to export all output of the checked projects together with the Java source files and resources, restrict the export to the listed resources, select a target directory and accept the default JAR packaging options. On the JAR manifest specification screen it prescribes generating a MANIFEST file and setting the main class main.MainSystem before finishing the wizard, which produces a JAR that starts the LifeSocial client directly. The instructions date from the phase before the platform was converted to OSGi bundles, when the lifecycle library "LifeSocial core" still had to be shipped as a plain, self-contained library.

HowTo (Studienarbeit-Begleitdokument, KOM – Multimedia Communications Lab, TU Darmstadt) App & Kernel PDF · BibTeX
2010 LifeSocial.KOM – A P2P-based Platform for Secure Online Social Networks (Demo-Poster, IEEE P2P 2010)
Kalman Graffi, Christian Groß, Patrick Mukherjee, Aleksandra Kovacevic, Ralf Steinmetz (Multimedia Communications Lab, TU Darmstadt)
abstract

This one-page demo poster accompanies the IEEE P2P 2010 demonstration of LifeSocial.KOM and presents it as a fully distributed, plugin-based P2P platform for secure online social networks. It lists what the demonstration shows: profiles, friends, groups, photos, messaging and chatting, secure communication and access-controlled storage, and integrated monitoring of the quality of service. The architecture part explains the layered platform, with FreePastry as the structured overlay and PAST as the replicated storage layer, secure storage and message dispatchers, an information cache, mandatory and optional plugins and a customizable, extendable GUI, following a GNU/app philosophy in which plugins define the functionality and their dependencies, with profile functionality mandatory and extras such as TicTacToe optional. The data part shows the social data structures in the DHT, distributed linked lists of small objects addressed by hashed storage keys such as H("my data"), and the communication principles are plugin-to-plugin communication with the PeerID equal to the user's public key for authenticated, encrypted exchange, plus ID-based location and distributed linked lists for shared storage. The outlook names autonomous management of the quality of the running system, security extensions such as deletion enforcement and detection of illegal content, additional interaction plugins for games and a whiteboard, and deployment in an alpha/beta test as an open source project.

Poster (Demo-Poster zur IEEE P2P 2010-Demonstration) Overlay (Pastry) Storage & DSS Messaging Monitoring Security Search Plugin system App & Kernel PDF · BibTeX
2010 LifeSocial.KOM: A P2P-based Platform for Secure Online Social Networks (Demo-Paper, IEEE P2P 2010)
Kalman Graffi, Christian Groß, Patrick Mukherjee, Aleksandra Kovacevic, Ralf Steinmetz (TU Darmstadt, Multimedia Communications Lab / KOM); gefördert durch DFG-Forschergruppe 733 "QuaP2P"
abstract

Friends List Friends Graph Photo Albums Image Browser In Postbox Out Postbox Compose Area Chat Window Whiteboard Schedule Poll File Transfer Monitoring Login Profile Groups Search Calendar Online social networks with millions of users are very popular nowadays. They provide a platform for the users to present themselves and to interact with each other. In this paper, we present a totally distributed platform for social

Publikation (vierseitiges Demo-/System-Paper, IEEE P2P 2010) Overlay (Pastry) Storage & DSS Messaging Monitoring Security Search Plugin system App & Kernel PDF · BibTeX
2010 Monitoring and Management of Peer-to-Peer Systems
Kálmán György Graffi (Erstreferent: Prof. Ralf Steinmetz, Korreferentin: Prof. Carmen Guerrero López)
abstract

This doctoral dissertation (Dr.-Ing., Technische Universität Darmstadt, submitted 26 May 2010, defended 20 July 2010) systematically develops decentralized monitoring and management mechanisms for structured peer-to-peer systems, addressing the problem that in a network of autonomous and unreliable nodes there is no central point at which the state could be observed and the behaviour controlled. SkyEye.KOM is engineered first: it builds a tree of monitoring domains deterministically from the peer identifiers of the underlying KBR overlay in a unified [0,1] identifier space that is mapped back onto the overlay space, maintains that tree solely with protocol messages, and aggregates at every node count, sum, minimum, maximum and sum of squares so that the update size stays independent of the number of samples while yielding mean, variance and standard deviation for the whole system; individual peer load limits are respected through support peers. In addition to the global status snapshot it provides capacity-based peer search, for example for five peers with a given upload bandwidth and free disk space. P3R3O.KOM then realizes reliable long-term resource reservation under churn by automatically adjusting the redundancy of the resource provisioning, the redundant peer assignment variant reaching an adjustable guarantee of 100 per cent from four redundant peers, while SkyNet.KOM implements the monitor-analyze-plan-execute cycle of autonomic computing so that preset quality intervals, such as a Chord hop count between 7 and 10, are reached and kept by systematically re-configuring parameters such as the finger table size. The evaluation combines analytical models, PeerfactSim.KOM simulations with up to 10000 peers, network delays and KAD-based churn, and a testbed of 500 peers on 37 machines, reporting information freshness better than about 200 seconds, monitoring overhead in the order of 100 to 140 bytes per second per peer and around 3 KB/s with a five-second update interval, capacity queries answered after some five to six hops, and stable results even under 50 per cent peer failure. LifeSocial.KOM, the platform for online social networks and predecessor of LibreSocial, integrates the mechanisms with FreePastry and PAST, distributed data structures, cryptographic access control and a wide set of plugins, and thereby demonstrates the feasibility and application scope of the monitoring and management solutions.

Dissertation (Dr.-Ing., TU Darmstadt, eingereicht 26.05.2010) Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2011 Adaptives Loadbalancing für strukturierte Peer-to-Peer-Netzwerke am Beispiel von Chord
Philip Wette (Matrikel 6382629); betreut von Prof. Dr. Christian Scheideler und Dr. Kalman Graffi (Universität Paderborn / HST)
abstract

This master's thesis develops and simulates adaptive load balancing for structured peer-to-peer networks, using Chord as the example, and argues that the deterministic assignment of keys to peers in a DHT ignores peer capacities, file sizes and request popularity, so that under Zipf-like access patterns individual peers become overloaded while resources elsewhere remain unused. After reviewing unstructured overlays such as Napster and Gnutella, structured overlays and Chord with its finger table, routing and stabilization, the limitations of Chord are analysed, including the missing self-stabilization and the variant Re-Chord, the resulting distribution of keys and responsibilities, Zipf-distributed requests with p(x) proportional to (1/rank)^0.6, and existing load-balancing and caching approaches. The own approach, Adaptive-Chord, decouples the serving of a file from the responsibility for it: load is defined as the data actually sent per unit time and overload relative to the upload capacity, a second Capacity-Chord overlay sorts peers by their free resources with identifiers derived from a load index plus a random tie-breaker, and an overloaded peer mirrors its most demanding file to a peer with free capacity and forwards requests with a redirect at the cost of one additional hop. The evaluation in PeerfactSim.KOM with 1000 peers, 10000 files of 0 to 10 MB and Weibull-distributed churn shows the average download time falling from 806.7 s with Chord to 85.9 s with Adaptive-Chord without churn, the success rate rising from 71.8 to 97.1 per cent, the download speed rising from 118.6 to 275.4 KB/s, and the number of overloaded peers dropping from 158.6 to 18.2, while with churn the download time is roughly halved, for example from 514.4 s to 186.1 s. As a secondary result the thesis evaluates the self-stabilizing Chord variant Re-Chord for the first time under realistic churn and finds it comparable to Chord without churn but collapsing after about two hours because of its O(n log n) stabilization rounds and the resulting message traffic, recommending replacement predecessor and successor pointers instead.

Masterarbeit (Dateiname als "Stud" abgekürzt) Overlay (Pastry) Storage & DSS Messaging Monitoring PDF · BibTeX
2011 LifeSocial.KOM: A Secure and P2P-based Solution for Online Social Networks
Kalman Graffi, Christian Groß, Dominik Stingl, Daniel Hartung, Aleksandra Kovacevic, Ralf Steinmetz (TU Darmstadt, Multimedia Communications Lab; teils Norwegian Information Security Laboratory, Gjøvik University College)
abstract

This paper presents LifeSocial.KOM, a peer-to-peer platform for secure online social networks that provides the functionality of common server-based services in a totally distributed manner and thereby shifts the operational load and cost from a single provider to the participating users. The core of the platform is the structured overlay FreePastry with PAST as the reliable, replicated storage layer; on top of them a storage dispatcher, a message dispatcher, an information cache and a plugin layer implement profiles, friend lists, groups, photo albums, messaging, live chat, status updates and collaboration tools such as file exchange, a whiteboard and games, with plugins that build on one another. Data are organized as distributed linked lists of small ID-addressed objects, which spreads storage and retrieval load over many peers. A security layer provides authenticated and confidential communication and decentralized registration and login by using the user's public key as nodeID and userID, together with access control on stored items: sensitive objects are encrypted with an object-specific symmetric key whose copies, wrapped with the public keys of the privileged users or of the group members, are attached to the signed CryptedItem. To make the quality of the distributed system visible to users and operators, the monitoring solution SkyEye.KOM for structured overlays is integrated and measures metrics such as data retrieval times, peer count and message traffic on several layers. An enterprise-sized testbed demonstrates feasibility and load distribution, with disk usage around 150 KB during stabilization that grows afterwards with the replication needed for data availability, only 2.1 to 2.5 KB/s upload and download bandwidth with compression, up to one message per peer and second, and neither peer count nor peer online time influencing the message overhead noticeably.

Publikation (IEEE CCNC 2011) Overlay (Pastry) Storage & DSS Messaging Monitoring Security Plugin system App & Kernel PDF · BibTeX
2011 LifeSocial.KOM: A Secure and P2P-based Solution for Online Social Networks
Kálmán Graffi, Christian Groß, Dominik Stingl, Daniel Hartung, Aleksandra Kovacevic, Ralf Steinmetz (TU Darmstadt, KOM / QuaP2P DFG-Forschergruppe 733)
abstract

These are the accompanying slides of the IEEE CCNC 2011 presentation of the paper on LifeSocial.KOM, given by Kalman Graffi for the Multimedia Communications Lab and the DFG research group QuaP2P. The talk opens with a provocative motivation: instead of the usual claim that client/server is bad, it argues that classical P2P applications such as file sharing, Skype, Wuala and Joost are losing ground to the cloud, so P2P on user devices survives only with controlled quality and new applications, and identifies online social networks as the most promising one. It contrasts what users want, storing and searching content, user-to-user interaction and security of their data, with what providers want, many users, low operational costs of roughly one dollar per user and year, and a controlled quality of service, and proposes using the capacities of user devices, in view of Moore's law, as in LifeSocial.KOM, SafeBook and PeerSoN. The architecture slides show the layered stack of the platform, user interface, optional and mandatory plugins, information cache and monitoring, secure storage and message dispatchers, storage and replication, and the overlay on top of standard Internet protocols, with FreePastry and its PAST layer chosen instead of academic or homegrown overlays, distributed linked lists as data structure, public keys as node identifiers, and symmetrically encrypted objects whose symmetric keys are attached in encrypted form for the privileged readers. SkyEye.KOM is presented as the integrated, fully distributed and cheap monitoring solution that reports global statistics such as CPU and bandwidth usage, retrieval delays, messages sent and received, peer count, cache contents and clustering coefficient as average, minimum, maximum and standard deviation. The deck concludes with a summary of the operational, secure and monitorable prototype, next steps such as integrating management mechanisms and running an Internet-wide beta test, and backup slides on academic issues such as the effort of engineering a prototype and the difficulty of testing large-scale distributed systems.

Vortrag (Foliensatz; Dateiname „.pdf“ enthält faktisch eine PowerPoint-97-Datei, Textextraktion daher über LibreOffice-Konvertierung) Overlay (Pastry) Storage & DSS Monitoring Security Search Plugin system App & Kernel PDF · BibTeX
2011 Monitoring and Managing Quality in P2P-based Online Social Networks
Kálmán Graffi (Dr.-Ing., KOM – Multimedia Communications Lab, TU Darmstadt / Universität Paderborn; Gruppen QuaP2P, DFG-Forschergruppe 733)
abstract

These slides document an invited talk about monitoring and managing the quality of peer-to-peer systems, given by Kalman Graffi in the group of Prof. Christian Scheideler on the theory of distributed systems, with the platform LifeSocial.KOM as demonstrator. The introduction names the challenges for the quality of P2P systems, namely their distributed character, an undefined scale from tens to millions of peers, churn, peer heterogeneity and static configurations that cannot cope with them, and argues that P2P on user devices can survive the cloud only by offering controlled quality together with new applications. The first contribution, SkyEye.KOM, decouples monitoring from the specific overlay through a unified identifier space [0,1], builds a tree of information domains whose domain identifiers are computed deterministically from peer identifiers, collects local measurements periodically, aggregates count, sum, minimum, maximum and sum of squares at each level and disseminates the global view from the root in O(log_beta(N) times the update interval. Its evaluation combines an analytical model, PeerfactSim.KOM runs with 10000 peers, network delays and KAD-based churn, and a testbed of 500 peers on 37 machines, showing precise values, an information age of about 200 s, a balanced traffic overhead of 100 to 140 bytes per second per peer, about 3 KB/s with a five-second update interval, and no critical effect even of a 50 per cent peer failure. SkyNet.KOM closes the management loop with an autonomic-computing cycle in which the root detects a violation of a predefined metric interval, plans a step-wise parameter change and spreads the new configuration in the acknowledgements of the monitoring updates, acting only when the slope of the metric history is small in order to avoid configuration oscillation; in Chord the hop count target of 7 to 10 is reached in two steps starting from a finger table of 20 and in three steps starting from 160, with corresponding effects on the lookup delay. The second half and the backup slides present LifeSocial.KOM as a secure, distributed and controlled online social network, with its layered architecture, FreePastry and PAST, distributed linked lists with obvious storage keys, cryptographic access control and integrated monitoring, and conclude that the mechanisms apply to any metric and parameter and use p2p overlays off the shelf.

Vortrag (Invited Talk, Karlsruhe; Foliensatz mit Backup-Folien) Overlay (Pastry) Storage & DSS Messaging Monitoring Security Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2011 Seminararbeiten 2011: "A Distributed Framework for Social Networks" (Sammelband, UPB Paderborn)
Sammelband aus Themenübersicht und fünf Seminararbeiten der Projektgruppe, betreut von Dr.-Ing. Kalman Graffi (Universität Paderborn): Kalman Graffi (Overview), Kavitha Jagannath, Pascal Kallien, Daniel Breitlauch, Jonathan Schlußler, Andreas Volk
abstract

This paper presents the seminar topic for the project group (PG) ”‘A Distributed Framework for Social Networks”’. There are 12 topics, that will be assigned to the participants of the project group. In the following, you find a short motivation of each topic and a list of research questions that need to be addressed in your paper. Please keep in mind, that the PG aims at developing a p2p-based framework for social networks. Several questions how this goal is aimed at best are open and need to be clarified through the commitment of your seminar paper. As an input for your seminar, a few papers are listed. They mark the initial stepping stone for further investigations. All papers should be downloadable from the university network.

Seminararbeit (Sammelband / Projektbericht) Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2012 A Distributed Framework for Social Networks — Final Report (Project Group 2011)
Andreas Volk (Teamleiter), Daniel Breitlauch, Kavitha Jagannath, Pascal Kallien, Jonathan Schlüßler — Universität Paderborn, Betreuer: Dr.-Ing. Kalman Graffi
abstract

This is the final report of the University of Paderborn student project group (project group 2011, supervised by Dr.-Ing. Kalman Graffi), which built a generic peer-to-peer framework on top of FreePastry and ported the distributed social network LifeSocial.KOM from direct use of the overlay to that framework. The report states the goals and the non-functional requirements (availability, performance, heterogeneity, security), reviews the FreePastry DHT routing and the PAST replication the framework builds on, and describes the framework concepts: push-based communication over named channels instead of polling (MessageChannel 1:1 and 1:N, TopicChannel N:M, AggregateChannel, StreamChannel), publish/subscribe with Scribe, streaming with SplitStream, security, user management, distributed data structures, aggregation and monitoring, and a test plugin. Security is provided by elliptic-curve public-key cryptography for signing, key exchange and key distribution plus 128-bit AES for content, with access control lists and the symmetric group keys attached in encrypted form to the secure storage items. New distributed data structures (distributed set, distributed linked list and prefix hash tree with bucket size 5) replace the loss-prone earlier storage; every group owns its own distributed linked list, which the new wall plugin uses intensively, and the chat plugin was converted from polling to push operation. Testing is documented in detail, with roughly 90 unit tests, all of them successful, work-station tests, and a pool-area test on 24 February 2012 with eight PCs (Core i7-2600, 8 GB RAM, openSUSE 11.4, Gigabit, Java 7) plus a separate bootstrap server and test-plugin master, after earlier attempts with five firewalled Windows PCs and forty Linux PCs with an automatic shutdown at 8 p.m. had failed; the report closes with the presentation of the prototype at CeBIT 2012 and with open issues: no NAT support and therefore only public IP addresses, the need to disable the Windows firewall, no node-ID collision check on storage access, no support for heterogeneous or mobile nodes, and FreePastry having had no active development since 2009 while being vulnerable to Sybil, eclipse and routing attacks; a planned HTML5/AJAX GUI with HTML5 sockets was to decouple the backend from mobile clients.

Projektbericht Overlay (Pastry) Storage & DSS Messaging Monitoring Security Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2012 Seminararbeiten 2012: "A Peer-to-Peer Framework for Social Networks" (Sammelband, UPB Paderborn)
Sammelband mit 16 Seminararbeiten der Projektgruppe, betreut von Dr.-Ing. Kalman Graffi (Universität Paderborn): Bahareh Mousavi Jazayeri, Jens Janiuk, Payman Alavi, Alexander Mäcker, Kangping Liu, Alexander Morast, Yevgen Mexin, Marcel Sander, Hongying Zhao, Deepak Chandra, Chintan Parekh, Anastasia Petkau, Subramanya Kuduvally Gurumurthy, Shyam Sandeep Polineni, Luqman Ahmad, Lars Bremer
abstract

1.1 Definition and history BitTorrent is a file sharing protocol, which is based on Peer- to-peer protocols are playing more and more im- peer-to-peer technology. It is used for distributing large files portant roles in the Internet file sharing world. BitTorrent or data over the internet [2]. BitTorrent is designed by Bram is one of the most popular protocols that provide robustness Cohen in April 2001. The first release was on July 2. 2001. and the resource utilization. Compare with the traditional From 2010 BitTorrent dominates the P2P in most districts server-client mode, user can get more efficient transmission of the world, while P2P dominates the upstream traffics. rate with BitTorrent. That is why it has started to domi- In 2011 from the report of sandvine, BitTorrent dominates nate internet traffic of the world and been widely used. This both Upstream and downstream traffic in Europe. But in paper describes the definition and the deployment of BitTor- north and South American, Ares and HTTP still dominate rent, relevant approaches, algorithm and some variations of the downstream traffic. As of January 2012, BitTorrent has BitTorrent. As conclusion, there will be some advice for the 150 million active users according to BitTorrent, Inc. The Project group. monthly users can be more than a quarter billion [9] [2].

Seminararbeit (Sammelband / Projektbericht) Overlay (Pastry) Storage & DSS Messaging Monitoring NAT traversal Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2013 A P2P-Framework for Social Networks
Kálmán Graffi (Jun.-Prof. Dr.-Ing.); im Dokument zitierte Mitautoren u. a. C. Groß, D. Stingl, D. Hartung, A. Kovacevic, R. Steinmetz, P. Mukherjee, B. Menges, L. Bremer, P. Wette
abstract

These 101 slides give a complete architecture overview of the P2P-Framework and the LifeSocial platform as it stood in 2013, presented by Jun.-Prof. Dr.-Ing. Kalman Graffi in the context of the CeBIT appearances ("See you at CeBIT 2013"). After a short history (development since 2007, more than 55 students involved) and the three goal perspectives of users, operators and researchers, the deck walks through the layer model of GUI framework, plugins and apps, P2P framework, overlay and Internet, and then through the modified FreePastry overlay: release of 13 March 2009 with 160-bit node IDs and base b = 4 giving O(log_{2^b} N) lookup hops, routing-table entries kept as buckets of k contacts for parallel routing and node replacement, iterative routing with alpha parallel lookups, failure recovery over the leaf set and routing-table order, and weak nodes recognised by port parity and carried only in the leaf set; since 2012 the FreePastry package is part of the framework source and can therefore be modified directly. The root of trust is the authenticated node ID, a 160-bit elliptic-curve public key hashed from user name and password, with immutable user IDs mapped to entries at hash(UserID), signed and encrypted communication (optionally with symmetric pairwise keys established for example by Diffie-Hellman), stored items changeable only by their original author, and 128-bit AES for content. Storage and replication extend PAST with in-place updates and with delivery by earlier downloaders for load balancing, add access control through user lists and nested groups with group keys, and host the distributed data structures (list, set, group and prefix tree), while communication covers one-to-one messaging, Scribe publish/subscribe and SplitStream-based aggregation. Four implemented search engines (local, network, Scribe and BloomJoin engines) are compared: the Scribe engine has the best query latency and copes well with many keywords, BloomJoin does not significantly reduce the network cost, and already 5 per cent churn hurts performance noticeably. Test results from runs with 25, 50 and 100 nodes report more than 19,000 photos in 13 minutes, 9,600 group messages in 8 minutes and 9,600 wall posts in 7 minutes, bandwidth per node always below 30 kb/s and about 3 kb/s on average, a replication factor of at least 2 and an average contribution of about 200 MB per node with a max-to-min spread of 350 MB to 50 MB; documented defects include a FreePastry memory leak (fixed), a deadlock in the message dispatcher that led to deactivating the chat plugin, wrong reuse of the login item after a crash, endless loops in the search plugin and a ClassCastException in the wall plugin under churn.; the final slides cover the monitoring, test plugin and app market services, the outlook on live multimedia streams, the CeBIT 2014 appearance and the planned open-source publication of the framework, and a list of related publications.

Vortrag (Übersichtsvortrag / Foliensatz, 101 Folien, CeBIT-Kontext) Overlay (Pastry) Storage & DSS Messaging Monitoring Security Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2013 A Peer-to-Peer Framework for Social Networks — Dokumentation (Abschlussbericht der Projektgruppe 2012, UPB Paderborn)
Payman Alavi, Lars Bremer, Deepak Chandra Veena, Jens Janiuk, Liu Kangping, Subramanya Kuduvally Gurumurthy, Ahmad Luqman, Alexander Mäcker, Yevgen Mexin, Alexander Morast, Bahareh Mousavi Jazayeri, Chintan Jayesh Parekh, Anastasia Petkau, Jan Marcel Sander, Hongying Zhao — betreut von Dr.-Ing. Kalman Graffi (Fakultät EI, Universität Paderborn)
abstract

This is the closing documentation of the University of Paderborn student project group of 2012 (fifteen students, supervised by Dr.-Ing. Kalman Graffi, Faculty of Electrical Engineering, Computer Engineering and Mathematics, report dated 30 April 2013), which developed LifeSocial, the predecessor of LibreSocial, on three levels at once: new Eclipse RCP plugins, extensions of the P2P framework, and modifications of the FreePastry overlay. After the status quo, the use cases of e-learning, collaboration and social networking, and the plugin and GUI challenges, the report presents the plugin work: the groups plugin implemented as an extension point to which other plugins register declaratively as IGroupContentContributor, an abstract browser based on jQuery, BBCode and string templates, forum, multi-view topic chat, file hosting, voting, file transfer and an app store built in four layers with the models PluginMeta, VersionMeta and PluginObjectFile and two private repositories, plus perspectives and auto-login in the GUI. Framework additions are the distributed tree, role-based access rights realised as hierarchically nestable groups whose keys are distributed by public-key encryption, write protection by signing the item together with a random nonce for replay protection and a four-step remote-operation protocol with hash and version comparison, voting and the AbstractStorageItem; groups are encrypted symmetrically with the group access rights, and public and private groups are supported. The evaluation of the distributed list shows that remote operations halve the traffic and that bucketing improves it further, that consolidation lowers the share of deleted entries from as much as 0.3 to usually below 0.05, and that the split size is a trade-off with a value of about 20 working well in the tests. The overlay chapter covers security against Sybil, eclipse and routing attacks through k-buckets with k = 5 and parallel and iterative routing with an explicit source-node role, capacity awareness through provider lists, an LRU cache and the RedirectedLookupMessage configured with a 200 MB cache, provider list length 20 and replication factor 2, and the exclusion of weak and mobile nodes from routing tables and storage while they remain reachable through the neighbor set. Testing comprises unit tests and pool-area testing with monitoring: on a ten-machine pool the heterogeneity mode brought almost no gain (615 versus 624 successful photo downloads, minus 1.5 per cent), LifeSocial reached only about 200 KB/s per machine although the network offered 75 MB/s, objects above roughly 3.1 to 4.3 MB made the system unreliable, and a twenty-machine test (Core i7-2600, 8 GB RAM, openSUSE 12.3) over a two-hour test plan left open defects in the friends and wall plugins, while the app-store test with 20 machines and 100 publications of 1.5 MB gave implausible byte-per-second monitoring values, and the appendices provide a start guide, a plugin development guide and a test guide.

Projektbericht / Dokumentation (PG-Abschlussbericht) Overlay (Pastry) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2013 A Peer-to-Peer Framework for Social Networks — Project Group 2012/2013
Jun.-Prof. Dr. Kalman Graffi / Projektgruppe „P2P Framework for Social Networks“, Institut für Informatik (Fakultät EIM), Universität Paderborn (u. a. genannt: Pascal, Studienarbeit zu Testplänen)
abstract

StorageItem I common way for identifying objects (using UUID) I allows specification, storage and lazy-loading of related objects I identified by a keyword (e.g. „posts“) I to-many: distributed data structures I automatically delete dependent objects

Vortrag (Folien zur Projektgruppe 2012/2013) Overlay (Pastry) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2013 Adding Capacity-Aware Storage Indirection to Homogeneous Distributed Hash Tables
Philip Wette (Computer Networks Group, Universität Paderborn), Kalman Graffi (Technology of Social Networks Group, HHU Düsseldorf)
abstract

These are the 25 slides (including a backup slide) of the NetSys 2013 presentation by Philip Wette (Computer Networks Group, University of Paderborn) and Kalman Graffi (Technology of Social Networks Group, HHU Düsseldorf) of the paper on making homogeneous distributed hash tables capacity aware. The motivation section recalls peer-to-peer and DHT basics with Chord as the prototype and argues that load becomes unbalanced even when keys and peers are evenly distributed in the identifier space, because file sizes and file popularities are heterogeneous, a small group of files is responsible for the largest share of traffic, and the resources required for hosting a file are roughly file size times popularity. The approach decouples responsibility for a key from the supply of the data: a second, modified Chord overlay whose identifiers reflect the free resources of a peer, ID0 = (2^m - 1) * min(1, v / M), lets an overloaded peer address capable peers and place mirrors of its most demanding files there, so the system architecture consists of a data overlay plus a capacity overlay, with the load-balancing algorithm presented as a flow chart covering the three cases of creating a mirror, removing a mirror and recalling a mirror from a supplier. The evaluation was carried out with the simulator PeerfactSim.KOM using GNP-derived latencies and a last-mile bandwidth model, with 1000 peers over 24 hours, 10,000 documents of 0 to 10 MB, Zipf popularity, query intervals of 10 and 30 minutes, popularity changes after 10 and 60 minutes, and with and without churn, counting a download as failed once the data rate falls below 1 kbps and a peer as overloaded once more than 95 per cent of its resources are consumed. The results table for the busy, unstable and no-churn scenarios shows downloads up to nine times faster, success rates close to 100 per cent (97.1 versus 71.8 per cent) and far fewer overloaded peers, and with churn still shorter download times and higher data rates (379.8 versus 249.2 kbps). The conclusion is that homogeneous DHTs are not designed for heterogeneous files and that a parallel capacity overlay with mirror-based load balancing reduces download times clearly, even under churn; the backup slide covers the treatment of non-uniform identifiers in Chord (Chord#).

Vortrag (Foliensatz zum IEEE-NetSys-2013-Paper, 25 Folien inkl. Backup-Folie zu Chord#) Overlay (Pastry) Storage & DSS Monitoring PDF · BibTeX
2013 Adding Capacity-Aware Storage Indirection to Homogeneous Distributed Hash Tables
Philip Wette (Computer Networks Group, Universität Paderborn), Kalman Graffi (Technology of Social Networks Group, HHU Düsseldorf)
abstract

Typical structured overlays such as Chord, Pastry and Kademlia consider only homogeneous nodes with equal capacities, which does not resemble reality, and in practical use cases nodes get overloaded by storing popular data. This paper presents a general approach to enable capacity awareness and load-balancing capability of homogeneous structured overlays by separating the indexing of data from its storage, so that the responsible peer keeps the index entry while a supplier peer explicitly chosen for its free upload capacity hosts and delivers the file; randomised caching and virtual peers were rejected because in P2P online social networks such as LifeSocial every location of a copy has to be known when content is updated. To find such suppliers efficiently, a hierarchical second structured overlay is introduced beside the data overlay, in which the identifiers are derived from the free resources according to ID = min(2^m - 1, (2^m - 1) * v / M), which sorts peers automatically by capacity, while peers whose maximum capacity is below a threshold kappa stay out of the capacity overlay. Protocols are specified for the data overlay (file redirection, file access statistics and load balancing with local mirror removal, mirror creation and remote removal, running every four minutes with kappa = min(0.2 * vmax, 70 kbit/s), gamma = 0.2, epsilon = 0.1 * vmax and omega starting at 150 kbit/s and varied by plus or minus 20 per cent according to success) and for the capacity overlay (join and position updates only when the capacity changed by more than epsilon, in order to avoid synthetic churn), with the required monitoring obtained by push-sum style aggregation as in SkyEye. In the PeerfactSim.KOM evaluation with 1000 peers over 24 hours, 10,000 files of 0 to 10 MB, Zipf popularity with p(x) proportional to (1/rank)^0.6 and Weibull-distributed churn, the number of overloaded peers falls from 160.4 with Chord to at most 24.5 with Adaptive Chord, and from 158.6 to 18.2 in the busy, unstable, no-churn scenario, while the download speed rises by up to 232 per cent (118.6 to 275.4 kbit/s), the download time drops by up to a factor of nine (806.7 s to 85.9 s) and the success rate grows from 71.8 to 97.1 per cent without churn. Under churn the gain is smaller because peers that go offline lose their mirrors and renewed mirroring itself consumes capacity, yet Adaptive Chord is still about twice as fast (186.1 s instead of 514.4 s with Chord) and improves the success rate from 38.1 to 43.6 per cent, so that long-established and well-evaluated overlays such as Chord can be kept while content, even very popular content, is hosted by easily findable capable peers.

Publikation (IEEE-Konferenzpaper, NetSys/ICCCN-Workshop 2013; gefördert durch DFG-SFB 901) Overlay (Pastry) Storage & DSS Messaging Monitoring PDF · BibTeX
2013 Benchmarking a Distributed Framework for Social Networks
Kangping Liu (Matrikel-Nr. 6672590, Uni Paderborn); betreut von Jun.-Prof. Dr.-Ing. Kalman Graffi und Prof. Dr. Christian Scheideler
abstract

This master's thesis at the Faculty of Electrical Engineering, Computer Engineering and Mathematics of the University of Paderborn (Paderborn, 27 October 2013, supervised by Jun.-Prof. Dr.-Ing. Kalman Graffi and Prof. Dr. Christian Scheideler) benchmarks the distributed framework for social networks LifeSocial.KOM, consisting of the FreePastry overlay, the P2P framework and its plugins, by improving its monitoring and by designing and running four kinds of tests: synthesis test, random test, long-term test and alpha test. The SkyEye.KOM monitoring mechanism was found to deliver inaccurate or useless values under load, so it was extended with a configurable aggregation over the last combinedRounds at the root category, which smooths global curves such as the number of nodes in the network at the cost of freshness, and with new metrics for the data flow from the plugins through the distributed data structures, the information cache and the framework down to the FreePastry layer (for example FRAMEWORK_SENT/RECEIVED_PLUGIN, ..._DDS, STORAGE_DATA_*, MESSAGE_DATA_* and PASTRY_DATA_*) plus social metrics such as friends, albums and posts, whereas measurements directly inside PAST and Scribe had to be dropped because their delays prevented nodes from joining. In the synthesis tests all test cases pass with 25 instances, and scaling from 25 to 100 instances raises the framework and plugin data streams by a factor of four to five while store, retrieve and lookup times remain good and the boot time grows by only about 0.1 s at fourfold size, the main pressure coming from the distributed data structures because they require many message exchanges. The random tests with churn reveal the weaknesses: many nodes stuck because a node that had logged out logged back in through the previous bootstrap node instead of through the master, and the friendship-request function, in which each test slave pulls a randomly chosen name from the master by RMI, blocked many slaves, so that the 100-instance network dropped immediately to 23 nodes whereas the 25-instance network remained stable at about 14 of 25, while without logout the system performed at a high level. The first long-term test shows LifeSocial running for 1500 minutes, about 25 hours, without exceptions before the FreePastry priority transport layer reports a queue overflow with massive message loss, and memory usage grows continuously, a leak that was finally removed by deactivating the runtime visualisation of the monitoring data and by adding the OSGi command "monitor <metric>"; the optional alpha test with real users failed because of a lack of motivation, with only two accounts and one day of activity in two weeks, but the thesis documents the lessons learned, including the deployment with Xvfb on ten virtual machines, duplicated wall posts and lost profile pictures, and concludes that small networks cope better with churn than large ones and that a motivated user group, preferably consisting of developers, is a prerequisite for such a test.

Masterarbeit Overlay (Pastry) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2013 Benchmarking a Distributed Framework for Social Networks
Kangping Liu (betreut von Jun.-Prof. Dr.-Ing. Kalman Graffi und Prof. Dr. Christian Scheideler)
abstract

This master's thesis at the Faculty of Electrical Engineering, Computer Engineering and Mathematics of the University of Paderborn (Paderborn, 27 October 2013, supervised by Jun.-Prof. Dr.-Ing. Kalman Graffi and Prof. Dr. Christian Scheideler) benchmarks the distributed framework for social networks LifeSocial.KOM, consisting of the FreePastry overlay, the P2P framework and its plugins, by improving its monitoring and by designing and running four kinds of tests: synthesis test, random test, long-term test and alpha test. The SkyEye.KOM monitoring mechanism was found to deliver inaccurate or useless values under load, so it was extended with a configurable aggregation over the last combinedRounds at the root category, which smooths global curves such as the number of nodes in the network at the cost of freshness, and with new metrics for the data flow from the plugins through the distributed data structures, the information cache and the framework down to the FreePastry layer (for example FRAMEWORK_SENT/RECEIVED_PLUGIN, ..._DDS, STORAGE_DATA_*, MESSAGE_DATA_* and PASTRY_DATA_*) plus social metrics such as friends, albums and posts, whereas measurements directly inside PAST and Scribe had to be dropped because their delays prevented nodes from joining. In the synthesis tests all test cases pass with 25 instances, and scaling from 25 to 100 instances raises the framework and plugin data streams by a factor of four to five while store, retrieve and lookup times remain good and the boot time grows by only about 0.1 s at fourfold size, the main pressure coming from the distributed data structures because they require many message exchanges. The random tests with churn reveal the weaknesses: many nodes stuck because a node that had logged out logged back in through the previous bootstrap node instead of through the master, and the friendship-request function, in which each test slave pulls a randomly chosen name from the master by RMI, blocked many slaves, so that the 100-instance network dropped immediately to 23 nodes whereas the 25-instance network remained stable at about 14 of 25, while without logout the system performed at a high level. The first long-term test shows LifeSocial running for 1500 minutes, about 25 hours, without exceptions before the FreePastry priority transport layer reports a queue overflow with massive message loss, and memory usage grows continuously, a leak that was finally removed by deactivating the runtime visualisation of the monitoring data and by adding the OSGi command "monitor <metric>"; the optional alpha test with real users failed because of a lack of motivation, with only two accounts and one day of activity in two weeks, but the thesis documents the lessons learned, including the deployment with Xvfb on ten virtual machines, duplicated wall posts and lost profile pictures, and concludes that small networks cope better with churn than large ones and that a motivated user group, preferably consisting of developers, is a prerequisite for such a test.

Masterarbeit Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2013 Seminararbeiten 2013: "A Peer-to-Peer Framework for Social Networks" (Sammelband, UPB Paderborn)
Sammelband mit 16 Seminararbeiten der Projektgruppe, betreut von Dr.-Ing. Kalman Graffi (Universität Paderborn): Thomas Jacobs, Christopher Holomek, Rohan Sukumaran, Maitri Modasia, Florian Schulte, Navendu Barua, Eshant Tulsidas Gour, Christy Jovita Bibiana Arockiaraj, Kailaash Balachandran, Christopher Petri, Siddharth Kaneriya, Rapolu Manasa Veena, Weifeng Peng, Chander Prakash, Prince Singh, Bing Gao
abstract

Class This paper presents an overview of OSGi, discusses the architecture in detail, explains benefits of using OSGi from Public methods both developers and business point of view. It then presents some current OSGi framework implementations. This pa- per also explains what an OSGi bundles means and how to create that, what kind of metadata it contains. It also high- Non-public lights some existing OSGi repository. It then presents some methods important aspects in Decentralization in Peer-to-Peer (p2p) networks which are Decentralized Dynamic Code Manage- ment for OSGi, Decentralized Service Discovery and a sur- vey on resource discovery mechanism.

Seminararbeit (Sammelband / Projektbericht) Overlay (Pastry) Storage & DSS Messaging Monitoring Search NAT traversal Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2014 A Peer-to-Peer Framework for Social Networks — Final Report (Project Group 2013/14, LifeSocial)
C. J. B. Arockiaraj, K. Balachandran, N. Barua, B. Gao, E. T. Gour, C. Holomek, T. Jacobs, S. K. B. Kaneriya, M. Modasia, W. Peng, C. Prakash, M. V. Rapolu, F. Schulte, R. Sukumaran; Betreuung: Jun.-Prof. Dr.-Ing. Kalman Graffi (UPB)
abstract

Browser: The abstract browser provides the basic functionalities to render HTML pages and websites in LifeSocial. The abstract browser is also used for text formatting in forums and wikis. This text formatting is handled via BBCode, which is a markup language and offers protection against phishing attacks. For the generation of HTML content the abstract browser uses StringTemplate. StringTemplate is an easy to use templating language and is usable in Java.

Projektbericht (Final Report einer studentischen Projektgruppe) Overlay (Pastry) Storage & DSS Messaging Monitoring Security Search NAT traversal Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2014 A Peer-to-Peer Framework for Social Networks — Project Group 2013/2014
Projektgruppe „P2P Framework for Social Networks“ 2013/2014, Institut für Informatik, Forschungsgruppe Theorie Verteilter Systeme, Universität Paderborn (Leitung: Kalman Graffi)
abstract

This closing presentation (with live demonstration) of the University of Paderborn project group "P2P Framework for Social Networks" 2013/2014, supervised by Kalman Graffi, summarises a year of work on the LifeSocial/P2P framework. On the overlay side the group introduced several separate rings, each maintained by its own ring manager (discover(), checkMessageHeader(), ring-ID and magic-number layer above a new transport layer) that stores discovered rings and caches the addresses of online nodes, so that nodes no longer have to know a fixed bootstrap node at login, and added NAT support through automatic detection of the external IP, port forwarding and relaying over "pilot" connections to public leaf nodes with a four-stage connection search (direct, own pilot, pilot of a leaf, Pastry routing). Two framework corrections are reported in detail: a replication bug in the (k,r) scheme, where updates to key #K left stale replicas standing because only the key was checked, fixed by invalidation and "dirty" marking, a replication manager, deletion from persistent storage and re-insertion; and a signing bug caused by Java serialisation producing different byte sequences at sender and receiver, fixed by transferring the already serialised byte array (transient data plus serializedData) instead of serialising again for signature verification. The graphical user interface was decoupled from Eclipse RCP: the instance now runs as a background service while the surface is a JavaScript single-page application in the browser (HTML5, AJAX, jQuery, Bootstrap, Knockback.js) using REST requests and WebSocket notifications, with multilingual support and responsive design. Almost all plugins were rewritten in encrypted, framework-compliant form (profile with per-field privacy, wall as news stream, notifications, messaging, chat with WebRTC audio and video, photos, search with Soundex pre-filtering and Levenshtein post-filtering, calendar, voting, groups, forum), and testing is described on three levels: unit tests, the distributed test plugin run on 20 machines in a master/slave arrangement with a new "sendFriendRequestInOrder" case, and weekly internal user tests on two public always-on nodes. The conclusion sees a solid, secure and practically usable framework and names the public alpha test, the final farewell to Eclipse RCP, an app store, video streaming, BitTorrent and further NAT strategies as next steps.

Vortrag (Abschlusspräsentation mit Demonstration) Overlay (Pastry) Storage & DSS Messaging Monitoring Search NAT traversal Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2014 Distributed Access Control Mechanisms for Peer-to-Peer Networks
Christy Jovita Bibiana Arockiaraj und Manasa Veena Rapolu (Masterarbeit im Verbund, betreut von Jun.-Prof. Dr.-Ing. Kalman Graffi und Prof. Dr. Christian Scheideler, Universität Paderborn)
abstract

This joint master's thesis by Christy Jovita Bibiana Arockiaraj and Manasa Veena Rapolu, supervised by Kalman Graffi and Christian Scheideler at the University of Paderborn, extends the groups/community plugin of LifeSocial by hierarchical communities and level-dependent access control, motivated by the use cases of e-learning and a social intranet. Previously the plugin supported only a single group level containing folders and files, with access control on that one level; the thesis introduces multi-level community hierarchies with parent and reference keys so that communities can contain inner-level communities. Access control is realised per level: members are granted either read or write rights, and only holders of write access or the owner may rename or delete an item, implemented through checkForWriteAccess and setUpdateAllowed. Storage goes through the StorageDispatcher with keys such as getGroupUserListStorageKey(group) and getUserGroupListStorageKey(userID); communities, folders and files are kept as AbstractStorageItem and GenericStorageItem and in a DistributedSet for members and DistributedLinkedLists for folders and files, and non-friends can be invited to a community by invitation and notification instead of being forced into a friendship. The browser front end is connected via the WebProvider plugin with RESTHandler, CommunityPluginHandler and GroupsPluginHandler, which return JSON to the JavaScript/MVVM interfaces built with knockback.js, jQuery, AJAX and Bootstrap, and the testing chapter documents the test plan with screenshots. The conclusion reports that deleting inner-level communities does not yet work - users must leave the top-level community and recovery of related content throws an exception - and names calendar sharing and forums inside communities as future work.

Masterarbeit Storage & DSS Plugin system App & Kernel PDF · BibTeX
2014 Secure Distributed Data Structures for Peer-to-Peer-based Social Networks
Jens Janiuk, Alexander Mäcker (Heinz Nixdorf Institute, University of Paderborn), Kalman Graffi (Department of Computer Science, University of Düsseldorf)
abstract

Online social networks are attracting billions of that allows to implement an efficient and fully retrievable nowadays, both on a global scale as well as in social enter- storage. Replication approaches like PAST [5] keep data in prise networks. Using distributed hash tables and peer-to-peer the DHT persistent, even when the initially hosting node fails. technology allows online social networks to be operated securely and efficiently only by using the resources of the user devices, Distributed data structures (DDS) such as lists, trees or sets thus alleviating censorship or data misuse by a single network are very popular constructs in the OSNs. Sets are used to operator. In this paper, we address the challenges that arise in implementing reliably and conveniently to use distributed data organize memberships in groups, to manage friends and to add structures, such as lists or sets, in such a distributed hash-table- entries to an interest set. Lists are used to maintain and order based online social network. We present a secure, distributed list a set of entries, e.g. in mailboxes, statuses and comments, or data structure that manages the list entries in several buckets albums. Trees maintain dynamically entries in categories, e.g. in the distributed hash table. The list entries are authenticated, allocating users in classes such as relationship status, age or integrity is maintained and access control for single users and also groups is integrated. The approach for secure distributed sex. They allow to easily search for profiles or categorized lists is also applied for prefix trees and sets, and implemented entries. These DDS are essential for the creation of a p2p- and evaluated in a peer-to-peer framework for social networks. based OSN, as they relieve programmers from the load to re- Evaluation shows that the distributed data structure is convenient implement the same logic for every single small application and efficient to use and that the requirements on secur

Publikation (Peer-Review-Konferenzpaper) Overlay (Pastry) Storage & DSS Messaging Security Search Plugin system App & Kernel PDF · BibTeX
2014 Secure Distributed Data Structures for Peer-to-Peer-based Social Networks
Jens Janiuk, Alexander Mäcker (Präsentierender, Heinz Nixdorf Institute, University of Paderborn), Kalman Graffi
abstract

This paper presents secure distributed data structures (DDS) — lists, sets and trees — operated on a DHT (FreePastry with PAST storage) for peer-to-peer social networks, addressing functionality, fine-grained write/read control, replicability and load balancing without central servers. Lists are split into buckets of configurable splitsize addressed by h(listName + floor(i/splitsize)), so a list of length n needs only n/splitsize lookups and load spreads over many peers; remote operations (get(i), set(i), contains(item) executed at the storage peer) nearly halve traffic, while message counts stay independent of splitsize. Write protection is two-stage: a bucket signature under the group key governs insertion, and an element signature plus stored nonce restricts modification to the author with replay protection; elements are symmetrically encrypted with per-user asymmetrically wrapped keys. Group items hold a symmetric key and an asymmetric pair stored encrypted per member; admin groups and nested groups (a member may itself be a group) derive access by a depth-first walk of the hierarchy, and removing a member regenerates the group key. Deletion inserts a synthetic deletedItem tracked in the list metadata, and a consolidation protocol shifts it exactly one bucket per invocation (two signatures, two storage peers, notification timeout) so no elements are lost. Simulation with 1000 list operations on 1 kB elements (setting A: 50 % write/delete, setting B: 7 % as in real OSN usage) shows consolidation reducing the deletedItem share from up to 0.3 to mostly below 0.05 at only slightly increased traffic, with splitsize around 20 a good trade-off.

Vortrag (Foliensatz, 19 Folien inkl. Backup-Folien) Overlay (Pastry) Storage & DSS Messaging Security Plugin system App & Kernel PDF · BibTeX
2014 Systematic Evaluation of a Decentralized Social Network (Folien zur Masterarbeit)
Dheeraj Balabhadruni, Weifeng Peng (Betreuung: Jun.-Prof. Dr.-Ing. Kalman Graffi, Prof. Dr. Christian Scheideler; Institut für Informatik, Theorie verteilter Systeme, UPB Paderborn)
abstract

This master's thesis by Dheeraj Balabhadruni and Weifeng Peng (University of Paderborn, 22 December 2014, supervised by Kalman Graffi and Christian Scheideler) evaluates the decentralised peer-to-peer social network LifeSocial systematically by means of its test plugin, monitoring layer and long-running experiments. Of the 26 test cases that existed in the test plugin only 7 still worked, because the 2013 p2p-framework project had invalidated 19 of them; these were deleted or rewritten, and further test cases as well as social metrics were added for practically all plugins, including groups, messaging, photos, voting, calendar and search. The scheduled synthesis test - a timed plan such as "5x TestCaseCreateCalendarEvent 2m" running for more than 130 minutes - was executed on 10 and on 20 instances and turned out functionally sound: no lost messages, join times below 0.4 s and short lookup times, with remaining problems mainly in the groups and wall plugins. The random test with churn remained problematic, losing more than 600 messages with 10 instances and over 2000 with 20 instances (about 0.5 per cent and 1.7 per cent of those sent) and producing hanging or crashed nodes; after standardising the ParameterGenerator so that test cases no longer depend on execution order and removing a five-second sleep, the number of exceptions dropped sharply but the message losses stayed at about 600 and 1900, whereas without churn no crashes or losses occurred. The metric set was cleaned and corrected - information-cache metrics removed, STORAGE_DATA_SENT/RECEIVED renamed to STORED/RETRIEVED and the plugin data-flow prefix changed from FRAMEWORK_ to STORAGEDISPATCHER_STORAGE_ - after which globally sent and received framework data were identical. The long-term test survived 5300 minutes (88.3 hours, roughly 3.67 days) instead of the 25 hours reached previously, but ended in a message-queue overflow caused by the PriorityTransportLayer of FreePastry, with a memory leak in the MonitoringPluginHandler (historySize reduced from 200 to 1) and a steadily growing memory demand of the overlay as remaining causes; because of these open problems a real-user test was not carried out.

Vortrag (Verteidigungsvortrag zur Masterarbeit) Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2014 Systematic Evaluation of a Decentralized Social Network (LifeSocial)
Dheeraj Balabhadruni und Weifeng Peng (betreut von Jun.-Prof. Dr.-Ing. Kalman Graffi, Prof. Dr. Christian Scheideler)
abstract

This master's thesis by Dheeraj Balabhadruni and Weifeng Peng (University of Paderborn, 22 December 2014, supervised by Kalman Graffi and Christian Scheideler) evaluates the decentralised peer-to-peer social network LifeSocial systematically by means of its test plugin, monitoring layer and long-running experiments. Of the 26 test cases that existed in the test plugin only 7 still worked, because the 2013 p2p-framework project had invalidated 19 of them; these were deleted or rewritten, and further test cases as well as social metrics were added for practically all plugins, including groups, messaging, photos, voting, calendar and search. The scheduled synthesis test - a timed plan such as "5x TestCaseCreateCalendarEvent 2m" running for more than 130 minutes - was executed on 10 and on 20 instances and turned out functionally sound: no lost messages, join times below 0.4 s and short lookup times, with remaining problems mainly in the groups and wall plugins. The random test with churn remained problematic, losing more than 600 messages with 10 instances and over 2000 with 20 instances (about 0.5 per cent and 1.7 per cent of those sent) and producing hanging or crashed nodes; after standardising the ParameterGenerator so that test cases no longer depend on execution order and removing a five-second sleep, the number of exceptions dropped sharply but the message losses stayed at about 600 and 1900, whereas without churn no crashes or losses occurred. The metric set was cleaned and corrected - information-cache metrics removed, STORAGE_DATA_SENT/RECEIVED renamed to STORED/RETRIEVED and the plugin data-flow prefix changed from FRAMEWORK_ to STORAGEDISPATCHER_STORAGE_ - after which globally sent and received framework data were identical. The long-term test survived 5300 minutes (88.3 hours, roughly 3.67 days) instead of the 25 hours reached previously, but ended in a message-queue overflow caused by the PriorityTransportLayer of FreePastry, with a memory leak in the MonitoringPluginHandler (historySize reduced from 200 to 1) and a steadily growing memory demand of the overlay as remaining causes; because of these open problems a real-user test was not carried out.

Masterarbeit Overlay (Pastry) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2014 Systematic Evaluation of a Decentralized Social Network (LifeSocial) — Masterarbeit mit vollständigem Anhang
Dheeraj Balabhadruni und Weifeng Peng (betreut von Jun.-Prof. Dr.-Ing. Kalman Graffi und Prof. Dr. Christian Scheideler, Universität Paderborn)
abstract

This master's thesis by Dheeraj Balabhadruni and Weifeng Peng (University of Paderborn, 22 December 2014, supervised by Kalman Graffi and Christian Scheideler) evaluates the decentralised peer-to-peer social network LifeSocial systematically by means of its test plugin, monitoring layer and long-running experiments. Of the 26 test cases that existed in the test plugin only 7 still worked, because the 2013 p2p-framework project had invalidated 19 of them; these were deleted or rewritten, and further test cases as well as social metrics were added for practically all plugins, including groups, messaging, photos, voting, calendar and search. The scheduled synthesis test - a timed plan such as "5x TestCaseCreateCalendarEvent 2m" running for more than 130 minutes - was executed on 10 and on 20 instances and turned out functionally sound: no lost messages, join times below 0.4 s and short lookup times, with remaining problems mainly in the groups and wall plugins. The random test with churn remained problematic, losing more than 600 messages with 10 instances and over 2000 with 20 instances (about 0.5 per cent and 1.7 per cent of those sent) and producing hanging or crashed nodes; after standardising the ParameterGenerator so that test cases no longer depend on execution order and removing a five-second sleep, the number of exceptions dropped sharply but the message losses stayed at about 600 and 1900, whereas without churn no crashes or losses occurred. The metric set was cleaned and corrected - information-cache metrics removed, STORAGE_DATA_SENT/RECEIVED renamed to STORED/RETRIEVED and the plugin data-flow prefix changed from FRAMEWORK_ to STORAGEDISPATCHER_STORAGE_ - after which globally sent and received framework data were identical. The long-term test survived 5300 minutes (88.3 hours, roughly 3.67 days) instead of the 25 hours reached previously, but ended in a message-queue overflow caused by the PriorityTransportLayer of FreePastry, with a memory leak in the MonitoringPluginHandler (historySize reduced from 200 to 1) and a steadily growing memory demand of the overlay as remaining causes; because of these open problems a real-user test was not carried out.

Masterarbeit (Fassung mit Appendix; Abgabe 12. Januar 2015) Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2015 Appstore for LibreSocial — A Repository based App-Management for P2P Framework
Maitri Modasia (Matrikel 6747719, Uni Paderborn); betreut von Prof. Dr. Christian Scheideler und Dr.-Ing. Kalman Graffi
abstract

This master's thesis by Maitri Modasia (University of Paderborn, title page 9 April 2015, supervised by Christian Scheideler and Kalman Graffi) develops an app store for LibreSocial: a repository-based app management with which users publish, share, search, download and install their own OSGi bundles as apps at runtime, without any central repository instance. Publishers create one or more repositories, share a repository by publishing its repository ID, and publish an app as a .jar together with metadata (name, description, category, version, keywords, app image) that they can edit or delete; users join foreign repositories and install or uninstall the apps listed there, described in the thesis as publisher and user roles with requirements for the repository, app, publisher view, shared repositories view and installed apps view. The framework layer keeps the bookkeeping in the overlay itself: PAST replication with the (k,r) scheme on the leaf-set nodes of the responsible peer, distributed data structures (distributed list, distributed set, prefix hash tree, distributed tree), encrypted buckets, user IDs that are cryptographic public keys, FreePastry routing with O(log N) effort for storing and retrieving, and a mapping item relating node ID to user ID, while responsible peers cache metrics such as memory and bandwidth. The graphical layer is a web-based single-page application in HTML5/CSS/Bootstrap with jQuery and Knockback.js in the MVVM pattern, completely decoupled from the back end and accessing plugin APIs only through the web-provider plugin with its embedded server and REST handlers, one plugin-handler class per plugin, JSON serialisation and WebSockets for back-end-driven events. Apps themselves are OSGi bundles whose manifest requires a Bundle-SymbolicName and declares exported and imported packages and an activator, and they pass the lifecycle installed, resolved, starting, active, stopping and uninstalled, with quality of service monitored by the standalone application SkyEye.KOM. Note on the source: the text extract contains only chapters 1 and 2 (introduction with requirements, and the LibreSocial and OSGi background), so the design, testing and results chapters are not represented and no concrete test figures are documented.

Masterarbeit Overlay (Pastry) Storage & DSS Plugin system App & Kernel PDF · BibTeX
2015 Appstore for LibreSocial — A repository based App-management for P2P Framework (Folien)
Maitri Modasia (Matrikel 6747719), betreut von Jun.-Prof. Dr. Kalman Graffi (Institute of Computer Science, Research Group Theory of Distributed Systems, Uni Paderborn)
abstract

Defence slides for Maitri Modasia's master's thesis (University of Paderborn, supervised by Kalman Graffi; slide date 30 May 2015), presenting the app store module of LibreSocial. After an overview of the server-free peer-to-peer architecture and a list of the available plugins - among them web provider, login, profile, friends, notifications, messaging, photos, groups, files, forum, voting, calendar, search, wall, chat, audio/video, community, app store, testing, monitoring, commands and error console - the slides motivate the module through the idea of an app market with third-party apps, repositories and app management. The core concept is repositories organised as folder structures: every person keeps a repository list, a repository is shared through its repoID together with the list of participating users, and the framework's distributed structures back it - a per-user repository list addressed by the reference key userID, a distributed linked list of app store repositories, distributed linked lists of the apps per repository as sub-items, and a distributed set of users per repository. An app is an OSGi bundle (.jar) plus metadata, installed and removed at runtime, with its contents kept as storage items in a shared drive rooted at the repository folder. Plugin and GUI are decoupled: the Java plugin offers an API through AppstorePluginHandler and getAppstorePlugin(), a REST handler returns JSON to a browser GUI built with HTML5, AJAX, jQuery, Bootstrap and Knockback.js, and three views are implemented - publisher view, shared repositories view and installed apps view - each split into repositories and apps with tabs, lists, modal windows and call-to-action buttons. Open points listed as future work are real app installation, a search function and richer metadata such as icon, screenshots and rating; the talk ends with a summary and a demonstration.

Vortrag (Folien zur Verteidigung der Masterarbeit) Storage & DSS Messaging Monitoring Plugin system App & Kernel PDF · BibTeX
2015 Data Synchronization and Local Data Access across Peer-to-peer Social Networks
Bing Gao (Matrikel 6643802, binggao@uni-paderborn.de); vorgelegt bei Prof. Dr. Christian Scheideler und Jun.-Prof. Dr.-Ing. Kalman Graffi (Fakultät für Elektrotechnik, Informatik und Mathematik, Uni Paderborn)
abstract

This master's thesis by Bing Gao (University of Paderborn, 8 May 2015, supervised by Christian Scheideler and Kalman Graffi) develops a synchronisation plugin for the peer-to-peer social network LibreSocial that lets a user share a local folder and keep it synchronised with a sync group without central servers. The problem is defined as two-way, one-to-one synchronisation with automatically exchanged source and target roles; in the framework only authorisation data - the sync group with user IDs and paths, stored as a SecureStorageItem - are kept, while the sync database itself is deliberately not replicated. The chosen algorithms combine the Merkle hash tree, one tree per folder, which quickly reveals changed, new and deleted files, with Rsync-style delta computation between file pairs, implemented with the Java library Jarsync through the class JRSync in the plugin and Hash.java in org.lifesocial.plugin.sync.merkle; file and folder paths with their hash values are maintained as entries of distributed linked lists, completed by SyncGroup.java and the message class MsgFileSync.java derived from AbstractBasicMessage. Invitations run through the notification plugin, joining happens by invitation or by request with a group ID, and large files are transferred through the framework in 500 kB blocks. The evaluation on five machines (three with Windows 7, one with Mac OS, one with Ubuntu) covers two and three peers, two sync groups per peer, files above 100 MB and above 1 GB, a list of 200 files, an empty folder and mixed operating systems, with every test case rated as good; files below 1000 kB are transferred quickly while larger ones settle at a constant rate of roughly 1.2 Mbps because of the 500 kB blocks and about three blocks per cycle, with cycle times averaged over five data sets for increasing numbers of peers. Future work addresses deltas larger than 500 kB, where Rsync becomes expensive in CPU time and simple copying would be faster, and conflict resolution when the same file is edited in parallel.

Masterarbeit Overlay (Pastry) Storage & DSS Messaging Monitoring Security Plugin system App & Kernel PDF · BibTeX
2015 Public testing of a peer-to-peer based social network application
Florian Schulte (Betreuung: Jun.-Prof. Dr. Kalman Graffi, Prof. Dr. Christian Scheideler; Heinz Nixdorf Institut / Institut für Informatik, Universität Paderborn)
abstract

This master's thesis by Florian Schulte (University of Paderborn, 2015, supervised by Kalman Graffi and Christian Scheideler) prepares and evaluates public user tests of LibreSocial: it documents which code changes were necessary to make the distributed peer-to-peer application usable by end users and what two successive test rounds reveal about stability, performance and actual usage. The state of the software is described layer by layer - FreePastry and PAST as overlay, the p2p framework with configuration, distributed data structures, access protection, monitoring and test simulation, the plugins and the web frontend - together with general changes such as the rename from LifeSocial to LibreSocial, installation and unattended operation. Two performance fixes proved decisive: monitoring output was moved from plain-text and CSV files to a SQLite database, so writing became fast and nearly constant in runtime, and the XML serialisation of the cache and persistence layers was replaced by the kryo library, so an object of 512 kilobytes is processed in about 2 ms instead of roughly 500 ms on the same machine, which lets LibreSocial run even on weak hardware such as a Raspberry Pi. The test setup consisted of ten virtualised bootstrap nodes in the university data centre (one Xeon core, 4 GB RAM, openSUSE 13.2, OpenJDK 1.8) plus an automated start and update shell script with remote control commands (new script, reset, exec), a monitoring pipeline with conversion tools, a configuration file, feedback channels and tasks for the users. The first test with doctoral students showed at most 14 connected nodes at 6400 minutes, about four days and ten hours, dropping to 11 and even falling below the ten bootstrap nodes for a period because the monitoring could not represent departed nodes historically and cumulative values therefore decreased; the tests produced many valuable bug reports, among them XSS and CSRF issues against the REST interface and the browser GUI, countered with a per-session secret in local storage instead of only a cookie and plaintext filtering, and faults in the automatic selection of the network interface. In the second test an open group of student groups exposed a critical fault in the signature checking of storage objects, in which rejected objects were resent permanently, causing an endless loop and extreme CPU load, so the author decided to disable signature checking on purpose at the cost of the security concept; only nine of roughly 20 to 25 users answered the questionnaire, four of them could not connect for technical reasons, the data-carrying plugins (wall, photos, files) were markedly more error-prone than pure messaging, and further user tests were not started because data were lost during multi-hour operation.

Masterarbeit Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2015 Public testing of a peer-to-peer based social network application (Folien zur Masterarbeit)
Florian Schulte (Betreuung: Jun.-Prof. Dr.-Ing. Kalman Graffi, Prof. Dr. Christian Scheideler; Institut für Informatik, Theorie verteilter Systeme)
abstract

Defence slides for Florian Schulte's master's thesis on public field testing of LibreSocial with real end users (Institute of Computer Science, Research Group Theory of Distributed Systems, University of Paderborn, supervised by Kalman Graffi and Christian Scheideler). The talk starts from the goals of stability, usability and the licensing and publishing questions of the project, presents the LibreSocial architecture with OSGi, Eclipse RCP and the HTML5 GUI, and reports the rename from LifeSocial to LibreSocial for legal reasons - with a survey about the new name, a new logo and a new website that served as download repository during the test phase. Two adaptations of the software are highlighted: the monitoring did not scale because the statistics files were read and rewritten completely every five seconds, so a SQLite statistics output replaced them, which made writing very fast and nearly constant in runtime at the price of larger files and a conversion step; and kryo serialisation instead of compressed XML reduced storing a typical 500 KByte file chunk from more than 500 ms to 2 ms while clearly lowering the memory footprint. Further slides cover GUI adjustments, a feedback tool with an integrated screenshot editor, and the test design with ten virtual machines at the university as permanently reachable bootstrap nodes, several test rounds with a growing number of users and Traq as bug and feedback management system with about 60 tickets in the database. The evaluation combines monitoring results before and after the changes, mail feedback, a survey and the bug reports: lost-message errors persisted after some days of operation because nodes could not process messages quickly enough and their queues overflowed; in the survey 33,33 per cent of the respondents could connect to a network directly while 11,11 per cent were completely unable to connect for technical reasons, messaging worked for only 11,11 per cent whereas most reported it as failing or never tried it, the photo gallery worked for 100 per cent, and the participants were also asked about interest in a Raspberry Pi version. The conclusion is that the system is usable well shortly after a network has been created but develops problems and monitoring inconsistencies over time, and that critical bugs - among them de-anonymisation, CSRF in wall posts and exceptions at logout - have to be fixed before more users are exposed to them.

Vortrag (Verteidigungsvortrag zur Masterarbeit) Overlay (Pastry) Storage & DSS Messaging Monitoring Search Plugin system App & Kernel PDF · BibTeX
2015 Testing of a Peer-to-Peer Framework-based Social Network
Siddharth Kaneriya (Matrikel-Nr. 6633319, Uni Paderborn); betreut von Jun.-Prof. Dr.-Ing. Kalman Graffi und Prof. Dr. rer. nat. Christian Scheideler
abstract

class of test case in LibreSocial contains three explicit implementa- tions: Preconditions, Parameters and OSGi commands. In order to execute certain test cases, we must fulfill some conditions related to that test case. While, executing test case, it will check whether it fulfills required preconditions or not. if preconditions are not being fulfilled, it will execute fallback. The fallback is a TestCase, that needs to be executed to fulfill the preconditions of the certain test case [14]. Parameters are required for the execution of test cases as parameters con- tains all the necessary information like (IDs, Contents, Names etc..) which provides input content to the test case. if the method for checking valid parameters return true, the test case will be executed or proceed towards next step. In case of missing or invalid paramater, it will generate all the needed paramaters. Successful execution of test case is only possible when all the preconditions are satisfied and all the available parameters are available. More information about implementaion fo test cases and test case structure can be found in the previous master thesis conducted. [15, 14].

Masterarbeit Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2015 Testing of a Peer-to-Peer Framework-based Social Network
Siddharth Kaneriya (Matrikel-Nr. 6633319, Universität Paderborn); betreut von Jun.-Prof. Dr.-Ing. Kalman Graffi und Prof. Dr. rer. nat. Christian Scheideler
abstract

class of test case in LibreSocial contains three explicit implementa- tions: Preconditions, Parameters and OSGi commands. In order to execute certain test cases, we must fulfill some conditions related to that test case. While, executing test case, it will check whether it fulfills required preconditions or not. if preconditions are not being fulfilled, it will execute fallback. The fallback is a TestCase, that needs to be executed to fulfill the preconditions of the certain test case [14]. Parameters are required for the execution of test cases as parameters con- tains all the necessary information like (IDs, Contents, Names etc..) which provides input content to the test case. if the method for checking valid parameters return true, the test case will be executed or proceed towards next step. In case of missing or invalid paramater, it will generate all the needed paramaters. Successful execution of test case is only possible when all the preconditions are satisfied and all the available parameters are available. More information about implementaion fo test cases and test case structure can be found in the previous master thesis conducted. [15, 14].

Masterarbeit (Druckfassung) Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2016 Convex Hull Watchdog: Mitigation of Malicious Nodes in Tree-based P2P Monitoring Systems
Andreas Disterhöft, Kalman Graffi (Technology of Social Networks, Heinrich-Heine-Universität Düsseldorf; DFG-Grant GR4498 1-1 „OverlayMeter")
abstract

Tree-based monitoring systems such as SkyEye.KOM aggregate local metrics bottom-up in a monitoring tree and push the resulting global view back to all peers, which makes them an attractive target for malicious nodes that can distort the aggregate and thus the control loop built on it. The paper presents DOMiNo ("Detection Of Malicious Nodes"), a passive security extension that rates incoming monitoring data sets without producing monitoring traffic of its own. Instant mechanisms check structure and plausibility: the origin of a data set is verified against the identifier range expected for that position in the tree (forged IDs being prevented by using public keys as identifiers), nonces reveal faked acknowledgements, meaningless values are rejected through checks such as min <= mean <= max and sumOfSquares <= max^2 * count, and missing acknowledgements are tracked over several attempts. In addition, a collecting mechanism rates data sets as outliers with a modified Z-score computed from the median absolute deviation (falling back to the mean absolute deviation), a threshold of 3.5 defining a "convex hull" around the median, and evaluates the score over several time intervals before data sets are discarded under one of the policies discard none, discard safe or discard all. The work is organised around an attacker taxonomy (parent, child, random, random-acknowledgement and root manipulation, free-riding, as well as empty, upwards/downwards and do-nothing patterns) and evaluated in PeerfactSim.KOM on Pastry with 5000 nodes, no churn, 228 parameter combinations, branching factors of 8, 32 and 96, and 5, 10 or 20 per cent malicious nodes. Without protection the relative error of the global view grows to as much as 998 per cent, whereas DOMiNo limits it to about 9 per cent at a high manipulation factor and to 2, 6 and 18 per cent instead of 11, 24 and 53 per cent for a manipulation factor of 2 with 5, 10 and 20 per cent attackers, the best configuration being a branching factor of 96 with a Z-score threshold of 3.5 and only moderate overhead; being purely passive, DOMiNo cannot detect loops in the tree or leaf nodes that simply stop reporting.

Publikation (IEEE LCN 2016) Overlay (Pastry) Storage & DSS Messaging Monitoring Security App & Kernel PDF · BibTeX
2016 Distributed Data Structures Improvement for Collective Retrieval Time (Structure Cache)
Read Al-Aaridhi, Ahmet Yükseltepe, Tobias Amft, Kalman Graffi (Informatik, Heinrich-Heine-Universität Düsseldorf)
abstract

Distributed data structures (DDS) built on a DHT consist of chained objects that each hold an object ID, pointers and a payload, so retrieving a deep structure collectively is slow: the items of a reference chain have to be fetched one after another by recursive lookups, and the retrieval time grows linearly with the depth. The paper proposes a Structure Cache attached to the root object of a DDS that keeps the overlay keys of all deeper objects, turning the recursive chain retrieval into a single parallel retrieval (interface IStructureCache with CacheStructure(depth) and BuildStructureFromCache()). The cache is bound to the root object and replicated with it, can be kept current through update commands or fixed intervals, costs (C-1) overlay-key entries per structure, and falls back to the recursive retrieval when it has been lost; it is therefore unsuitable for structures that change very often. In a simulation with PeerfactSim.KOM, a Chord overlay, 1000 nodes, the GNP network model, KAD-style churn and a 100-entry deep tree, 1000 collective retrievals were measured with and without churn. The collective DDS retrieval time drops from roughly 190 s with churn and 165 s without churn to about 40 s and 30 s respectively, while the price is a higher time for a single item retrieval (about 1.1-1.4 s without cache versus 18.4 s with cache, because the cache has to be created once). Functional correctness is preserved: store and DDS retrieval succeed in 100 per cent of the cases and lookups in at least 99 per cent across all four scenarios, with about 400 contacted nodes and some 600 forwarded messages per node in the test scenario.

Publikation (IEEE WPMC 2016, Kurzpaper) Overlay (Pastry) Storage & DSS Search PDF · BibTeX
2016 Sets, Lists and Trees: Distributed Data Structures on Distributed Hash Tables
Raed Al-Aaridhi und Kalman Graffi (Heinrich-Heine-Universität Düsseldorf; DFG-Förderung GR4498 1-1 "OverlayMeter")
abstract

A distributed hash table only offers storage and retrieval of single items, which leaves P2P applications such as online social networks to manage their own complex data. The paper presents a Distributed Data Structures (DDS) layer between the application logic and the DHT in which sets, lists, trees and free graphs are realised as pointer graphs of ordinary DHT objects, each object consisting of an object ID, pointers and a payload; the entry point of a structure is either a deterministic hash or a random identifier, and the objects are replicated with PAST. Store, retrieve and modify operations (add, delete using tombstones) are defined for these structures, and the approach is contrasted with cooperative file systems, Ivy, PeerGraph, XML-Store, P2P-based social networks such as PeerSoN/OverSoc and LibreSocial. The evaluation uses PeerfactSim.KOM with Chord, PAST replication, the GNP network model and KAD-style churn in five scenarios: a photo-album example with 4 and with 11 items, a deep tree with 100 entries, a broad tree with 101 items over three levels, an average scenario with 200 DDS of 56 items each, and varying DDS sizes between 10 and 100 items, all with up to 1000 nodes. Success probabilities lie close to 100 per cent for stores, DDS retrievals, item retrievals and lookups (only the smallest scenario varies, with 94 per cent stores and 96 per cent lookups), hop counts stay between 5 and 6 throughout, store operations take 0.45-0.75 s except 4.28 s in the broad-tree scenario where 100 items are stored in parallel, and modifying a single item costs only 764 bytes and about 1149 ms. Broad structures clearly profit from the parallel retrieval, whereas deep structures suffer from the recursive chain of pointers and churn-related timeouts (a DDS retrieval of 130 s instead of 19 s without churn, and 313 s instead of 104 s in the larger setting), with roughly 400 contacted nodes per 100-item DDS and about 600 forwarded messages per node over 102,000 item retrievals; churn raises latency but hardly affects the success rates.

Publikation (IEEE-Konferenzpaper) Storage & DSS Monitoring Plugin system App & Kernel PDF · BibTeX
2017 Access Control for Secure Distributed Data Structures in Distributed Hash Tables
Raed Al-Aaridhi (Technology of Social Networks, Institute of Computer Science, Heinrich-Heine-Universität Düsseldorf); Projektgruppe DDS um Kalman Graffi
abstract

This poster, presented at IEEE LANMAN 2017 in Osaka, visualises the security architecture of the Distributed Data Structures (DDS) middleware for DHT-based P2P networks, whose goal is a security model without trusted third parties. It starts from the observation that P2P networks offer only storage and retrieval of single items, and shows how sets, lists and trees are built from pointer-linked data items that are distributed over the nodes and kept available by a replication protocol. The security model is illustrated with a distributed profile containing a photo album set, the pictures and a timeline entry with comments. A Security Context groups the security module of a DDS and consists of an Access Control Map, the Secured Data Items and a Configuration; access rights are recorded as signed Access Control Entries with subject key, entry type (ADD_RIGHT, REMOVE_RIGHT), access right (FULL_ACCESS, READ_ONLY), the data encryption key encrypted for the grantee, an encryption version, timestamp, signer and signature. Overlay keys are derived from the owner's public key, which makes ownership provable, dedicated key pairs allow anonymous identities that hide a grantee's real identity from third parties, and access removal works as lazy revocation with re-encryption only when data change. The demonstration setup is a list of 15 elements with several grantees holding different rights.

Poster (IEEE LANMAN 2017, Osaka, Juni 2017) Storage & DSS Security PDF · BibTeX
2017 Access Control for Secure Distributed Data Structures in Distributed Hash Tables (Short Paper)
Read Al-Aaridhi, Ahmed Yükseltepe, Kalman Graffi (Informatik, Heinrich-Heine-Universität Düsseldorf)
abstract

This short paper develops the access control mechanism of the Distributed Data Structures (DDS) middleware, addressing the question of how distributed structures such as sets, lists and trees can be kept confidential and integral in a public, untrusted DHT without any trusted third party. The assumptions are an existing public-key infrastructure, provable ownership of data, and potentially malicious storage nodes. Ownership is established by deriving the overlay key as OverlayKey = PublicKey(ObjectName), so every DHT node can verify who created a structure. Hybrid cryptography is used: the data of a DDS are encrypted once with a symmetric data encryption key (DDEKey), while that key is stored asymmetrically encrypted for each authorized user in an Access Control Map that follows the ACL pattern - a list of signed Access Control Entries (access right, encrypted DDEKey, signer) per public key, the creator's first entry proving the intended creation - so only a single copy of the encrypted data is needed; the approach is described as more suitable for larger data than purely asymmetric encryption. Secured Data Items form a map from DDS attribute names to encrypted items, an attribute-granular protection the authors call Granular Encryption/Granular Signature (GEGS), and a Configuration parameter, the encryption version, controls which DDEKey applies to new modifications. Revocation is solved by lazy revocation: a user whose entry has been removed may still hold the old DDEKey, so older versions of the data are deliberately ignored and only a new key, not a re-encryption of all data, has to be produced; anonymous identities based on dedicated key pairs additionally protect the real identity of grantees.

Publikation (Kurzpaper, IEEE LANMAN 2017) Storage & DSS Messaging Security App & Kernel PDF · BibTeX
2017 Evaluation of the Search Mechanism in LibreSocial
Moritz Kanzler (Heinrich-Heine-Universität Düsseldorf)
abstract

Searching is a crucial mechanism in a structured, decentralised social network, because the DHT layer itself only offers reliable key-value lookups for unique identifiers while range queries and attribute searches must be built on top of it. This Studienarbeit, written in the course "Opportunistic and Peer-to-Peer Networks", takes stock of the search mechanisms that are actually implemented in LibreSocial and compares them with the search functionality reported for other networks. A literature review distinguishes generic from application-specific search and defines five relevant categories - exact, keyword, substring, multi-attribute and range search - before the LibreSocial stack (FreePastry, the P2P framework and the OSGi plugins) and its building blocks, distributed sets and lists as well as prefix hash trees, are described. The implementation studied differs from the original "criteria list" concept by Podrajanski: today a join-query engine, Soundex n-grams for substring matching and a Levenshtein-distance ranking are used. The central finding is that only three of the five categories are supported, namely exact, keyword and substring search, while range search and multi-attribute search are missing, and that only profile data (user name, city and country) are indexed, so the mechanism is essentially a "user finds user" feature. An experimental analysis with a dedicated test plugin and SkyEye.KOM monitoring, 56 LibreSocial instances on seven PCs and two test scenarios, shows that adding photo albums, photos and messages does not increase the search-index data held in the overlay: the recorded curves stay constant on the order of a few kilobytes per node after the five-minute mark, and the TestCaseSearch covers only one of the three attribute categories. Known implementation flaws are reported as well - private profiles and blocked users are not filtered from the results, and a fixed five-second timeout per query makes the GUI react sluggishly - and the outlook discusses prefix hash trees, NetLocJoin and OptimizedRangeJoin, and Scribe publish/subscribe as ways to add range and multi-keyword search.

Studienarbeit Overlay (Pastry) Storage & DSS Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2017 LibreSocial Report – Summary of known Bugs (maybe outdated)
Patrick Szewior (Universität Paderborn, Tests/Alpha-Test)
abstract

This project report by Patrick Szewior (dated 30 March 2017 and explicitly marked as possibly outdated) summarises the experience gained from long-running tests and the alpha test of LibreSocial. The first part lists the fixes and tuning measures that were carried out: correcting the IP discovery in the test plugin, which failed on Debian and Ubuntu systems carrying several loopback addresses of the form 127.x.x.x; setting the java.rmi.server.hostname property at runtime so that test slaves can connect to the master via RMI; moving the LibreSocial storage and the monitoring database out of an NFS-mounted home directory, after Kerberos ticket expiry was believed to cause crashes twelve hours after start; replacing "==" with ".equals" in string comparisons; and changes to monitoring snapshots, the partitioning of monitoring databases, the tree splitting factor, delays between and before test cases, the DDS split size, the JVM heap size, the number of replicas, the hash calculation during registration and login, and the routing configuration. The second part collects known open problems: compile problems in Eclipse, logout not working, login reporting the user as not found, faults in the messaging and live-chat plugins, exceptions during the decryption of distributed data structures, crashes when loading https URLs from certificate authorities whose certificates the JVM does not accept, insufficient NAT, firewall and port-forwarding handling - servers with both a public and a private address being wrongly classified as weak, and manually configured forwarding being ignored in favour of a private address and external port 0 - and crashes of instances when monitoring is enabled, suspected to come from monitoring hash maps that are never cleaned up. The final part documents the configuration used for the alpha test of a network of 8 to 10 bootstrap nodes on one or two physical servers: hardware requirements (HP Moonshot X2150 servers with an AMD Opteron quad-core at 1.5 GHz, 8 GB RAM, 32 GB SSD), software requirements (a minimal desktop such as lxde-core, vncserver, xfonts-basic, firefox), network interface configuration, and the step-by-step setup of the first and of further bootstrap nodes, with monitoring switched off because of the unresolved crashes.

Fehlerliste / Projektbericht (Done Tasks, Known Bugs, Alpha-Test-Konfiguration) Overlay (Pastry) Storage & DSS Monitoring NAT traversal Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2017 SkyEye: A tree-based peer-to-peer monitoring approach
Kalman Graffi, Andreas Disterhöft (Heinrich-Heine-Universität Düsseldorf)
abstract

Peer-to-peer overlays that carry applications such as distributed online social networks have to adjust their system parameters in order to keep a promised quality of service, for instance maximum response times, and such control loops need a precise and fast-adapting picture of the current system state. SkyEye is presented as a decentralised, tree-based monitoring approach that runs on top of an existing structured overlay with key-based routing: every peer delivers local metrics, which are aggregated bottom-up and then disseminated proactively (push-based) to all peers. The monitoring topology is not maintained as explicit state but derived deterministically from the overlay identifier space: the unit interval is segmented recursively into domains with a branching factor beta and a depth of O(log N), the domain keys are resolved by the overlay, and the peers responsible for those keys act as coordinators with parents and same-level coordinators. Metrics are combined with commutative and associative aggregation functions (minimum, maximum, sum, count), the join and leave behaviour as well as churn are handled by coordinator replacement and same-level gossip, which prevents the global view from falling back to stale data. The evaluation in PeerfactSim.KOM identifies an update interval of 30 seconds and beta = 4 as the best compromise - an interval of 60 seconds visibly degrades precision while 15 seconds brings hardly any gain at double cost - and shows the monitoring layer costing only 512 bytes per minute and peer (about 1032 bytes per minute per peer in total), independent of the branching factor. Comparing overlays, Chord and Pastry yield the same monitoring quality, but Chord needs 4.6 hops per lookup and 26,294 bytes per minute per peer against roughly one hop and 1032 bytes per minute per peer for Pastry; against a push-sum gossip approach of the Gnutella 0.6 and Pastry kind, SkyEye reaches an absolute error of about plus or minus 0.1 instead of plus or minus 0.7, converges to the correct peer count within about two minutes and halves the monitoring traffic to 1 instead of 2 kB per minute per peer at comparable overall throughput. Under KAD-style churn on Chord the peer count is overestimated by at most 4 per cent, monitoring traffic rises to up to 103 per cent of the monitoring volume and network load by 12 to 26 per cent, with same-level gossip (MaxGossipLevel 7, MaxNoOfMsgs 32) keeping the results stable, so SkyEye outperforms push-sum in both precision and adaptation speed.

Publikation (Journal: Pervasive and Mobile Computing, Elsevier; DOI 10.1016/j.pmcj.2017.07.003) Overlay (Pastry) Storage & DSS Messaging Monitoring PDF · BibTeX
2018 Systematic Benchmarking of a Fully Distributed Communication System
Iakov Arkadjevic Dlikman (Betreuer: Jun.-Prof. Dr.-Ing. Kalman Graffi, M. Sc. Newton Masinde)
abstract

This thesis focuses on the evaluation and benchmarking of a peer-to-peer (p2p) based framework. The p2p-framework builds on top of the structured p2p-overlay Pastry. It offers a platform for distributed applications. These can be constructed using the modular OSGi environment and communicate with the p2p-framework through a well-defined application programming interface (api). The currently available application is LibreSocial, a distributed, online social network.

Masterarbeit Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2018 Systematic Benchmarking of a fully Distributed Communication System (Folien zur Masterarbeit)
Iakov Arkadjevic Dlikman (Institut für Informatik, Heinrich-Heine-Universität Düsseldorf)
abstract

These are the defence slides of Iakov Arkadjevic Dlikman's master's thesis "Systematic Benchmarking of a fully Distributed Communication System", presented at the Institut für Informatik, Heinrich-Heine-Universität Düsseldorf on 11 July 2018. The work asks for the capabilities and limitations of the P2P framework and of the distributed application LibreSocial in terms of quality of service in a small-scale local-area testbed, and adopts Kwok's metric definitions: availability pA = number of successful transfers / number of requested transfers, with scalability and robustness expressed as P(pA = 99 %) under a given population-growth or peer-dynamics model; the measurements are collected through the tree-based monitoring protocol, in which every peer sends measurements to its parent and receives the aggregated results, so that each peer gains sight of the whole network. In the availability scenario the storage availability drops as the number of action repetitions grows (photos 100 %, 92,5 % and 66,5 % in runs A, B and C, with only 34 % availability for uploading 128 photos; groups 89,25 %, 68 % and 52,25 %), while messaging behaves more stably (100 %, 98 % and 54 %) and forum thread creation fails almost completely. Under the exponential population-growth model the system keeps 86 % storage and 81 % messaging availability up to the 64 peers allowed by the physical setup, whereas the linear growth models show unexpected behaviour (as low as 24 % storage in LC); for robustness, light churn of 16,5 % (8 of 48 peers leave and rejoin) is tolerated with 75 % storage and 94 % messaging availability, but at medium churn of 33 % availability collapses to zero after the fourth round. The further analysis quantifies the framework layer itself: the storage dispatcher reaches data rates of about 100 KBps and the message dispatcher up to about 1,7 MBps with an average store time of 0,056-0,145 s, and with rising numbers of queries the storage layer limits the storage-heavy scenarios while the application layer limits messaging. An appendix reports a regression test that confirms the results of earlier work, a plugin map and a plugin-impact table identifying filestorage, photos and groups as the plugins dominating storage and messaging, and the qualitative caveat that the synthetic workload of 64 peers acting in lockstep is harsher than realistic user behaviour, while the conclusion and outlook recommend high-performance computing (up to 128 physical machines and more than 1024 LibreSocial instances), PlanetLab for realistic Internet conditions with NAT traversal and heterogeneous links, and full automation of the experiments through the OSGi console driven from Bash.

Vortrag (Verteidigungsvortrag zur Masterarbeit) Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2020 Benchmarking of a Distributed Framework for Online Social Networks
Liat Khitman (Betreuer: Newton Masinde M. Sc.; Arbeitsgruppe Jun.-Prof. Dr.-Ing. Kálmán Graffi, Heinrich-Heine-Universität Düsseldorf)
abstract

This thesis focuses on the evaluation and benchmarking of LibreSocial, a fully distributed and secure online social network. LibreSocial is built on top of FreePastry to enable secure routing between peers in the network. Furthermore, plugins could be developed and added to the system thanks to the utilization of the OSGi modular system development to increase LibreSocial’s quality service as a social media application. Prior to this thesis some work regarding the testing of LibreSocial was conducted, however little was done in terms of large scale testing to find out the functionality of the application in a large environment. In order to benchmark LibreSocial we utilize the integrated testing plugin. To gather the results we use a distributed monitoring tool, SkyEye.KOM, to gather statistics about different metrics in the application. We design three unique test scenarios with the aim to test several quality aspects of LibreSocial, namely stability, scalability, performance and cost. The tests are deployed to the high performance cluster at the university that provides capability to support up to 4000 LibreSocial nodes. We successfully show that LibreSocial is scalable, up to 2000 nodes, while remaining stable and during other test scenarios. We also show that LibreSocial maintains high performance under low as well as high workloads. Moreover, by varying the replication factor values to find an ideal value we show that LibreSocial performs well while maintaining low computational costs. We also indirectly show, as a result of some failed test attempts, that LibreSocial is relatively robust when internal failures occur.

Bachelorarbeit (Studienarbeit) Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2020 Caching Structures for Distributed Data Management in P2P-based Social Networks
Newton Masinde, Moritz Kanzler, Kalman Graffi (Heinrich-Heine-Universität Düsseldorf)
abstract

This paper presents and evaluates a social caching mechanism for the DHT-based distributed online social network LibreSocial (formerly LifeSocial.KOM), with the aim of reducing the load on the peers that store content and of shortening retrieval delays. Instead of the request-reply pattern over the modified FreePastry/PAST layer, content updates are pushed actively to a small set of selected "friends" over up to 15 parallel instance channels, so that data can be served from the local information cache with one-hop, predictive pre-loading; the set of monitored contacts is kept in a MUC list that is limited to the Dunbar number. Three cache-selection strategies are proposed and compared - random, trend, and a social-score strategy that combines tie strength with measures of the ego network's centrality - and they are measured with respect to performance, local efficiency and overlay efficiency. In the laboratory test with 64 LibreSocial instances on eight Debian machines, a downsampled Facebook'09 interaction trace, 15 social update channels, a MUC list of 150 users and a 50 s update interval, the three strategies achieve similar cache hit ratios of 94,9 % (random), 93,3 % (trend) and 92,4 % (social score), but the social score needs clearly fewer local resources, with the smallest social cache, the fewest subscription messages and an average memory use of 250-400 MB per instance, whereas random shows a continuously growing cache and subscription count. A second experiment over two days compares four settings - no cache, the existing time-window cache only, social cache only, and both together - and shows the best performance for the combination, with 99,2 % of requests answered from cache and only 0,8 % of requests reaching the overlay, against 94,5 % for the default cache and 84,9 % for the social cache alone. The price of this combination is a higher local storage demand and more overlay traffic from the active dissemination, and the paper notes that only the social cache keeps the cached content consistent when items are updated, while the time-window cache can serve stale data.

Publikation (IEEE ISNCC 2020) Overlay (Pastry) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2020 Capacity Management Protocol for a Structured P2P-based Online Social Network
Newton Masinde, Sebastian Bischoff, Kalman Graffi (Heinrich-Heine-Universität Düsseldorf, Technology of Social Networks)
abstract

Peer-to-peer (P2P) networks were first made popular by file- The growth of the Internet has been paralleled by the proliferation sharing applications such as KaZaA, Napster, LimeWire and BitTorrent. of computing devices with varying computing capabilities (such These early applications on P2P networks relied exclusively on always as storage, processing speed and bandwidth), from the traditional connected users to ensure that the services were available by utilizing some form of centralized indexing services. With the rapid technological desktop computers to much smaller, more mobile devices such as developments, the computing devices have become increasingly smaller laptops, palmtops and mobile phone devices. In many instances, and are connected via mobile networks run by teleecommunication these devices connect to the Internet via a mobile telecommunication companies. Additionally, more sophisticated applications that are even network. As a consequence, the Internet has evolved to support this more resource intensive such as online social networks (OSNs) on P2P platforms are possible. The smaller devices that are connected to the P2P heterogeneity. However, the heterogeneity exhibited in terms of com- OSNs may not have adequate resource capacities, such as bandwidth, puting capabilities of the interconnected devices presents a challenge memory or storage space, to share with the P2P network. In this paper, in P2P network, especially if the number of strong computing devices we present an algorithm that handles capacity management as he new are below a certain threshold. We differentiate between “strong” users join the network. We implement and test the algorithm on our P2P framework for OSNs called LibreSocial, and show that network functions and “weak” computing devices based on the resource capacity, with well even when there is a significantly high number of limited capacity strong devices having sufficient resources and weak devices lacking devices join the net

Publikation (IEEE OSN&T 2020, Kurzpaper) Overlay (Pastry) Storage & DSS Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2020 LibreSocial: A peer-to-peer framework for online social networks
Kalman Graffi (Honda Research Institute Europe), Newton Masinde (Heinrich-Heine-Universität Düsseldorf)
abstract

This open-access special issue paper gives the first complete description of LibreSocial, the peer-to-peer online social network that was developed since 2008 as LifeSocial.KOM, and derives from the cost, privacy and trust problems of centralised networks such as Facebook a catalogue of basic requirements a P2P-based OSN has to satisfy - identity and node identifiers, routing, communication channels, availability of data, distributed data structures, user and group management, security, monitoring, plugin extensibility and a graphical user interface. Related work on PeerSoN, Safebook, My3, eXO, MyZone, Cachet, DECENT, LotusNet, SuperNova and DiDuSoNet is compared with these requirements, and the architecture is then described layer by layer: a heavily modified FreePastry overlay, the P2P framework, OSGi plugins and applications, and the Web GUI. Identity rests on a 160-bit elliptic-curve public key as node ID (replacing the former 1024-bit RSA) and on an immutable user ID that is mapped to the current node ID through signed mapping items protected by a challenge-response mechanism, so that login works from any device with user name and passphrase only, while AES-128 protects content. Overlay and storage modifications include Kademlia-like buckets in the routing table for parallel, iterative routing, the marking and exclusion of "weak nodes" from routing and replication, and extensions of PAST with UPDATE and DELETE, store receipts verified by file hash, replica and file diversion for load balancing, plus distributed sets, lists and prefix hash trees with encrypted, signature-protected access control down to group level, SkyEye tree-based monitoring with about 600 metrics and the TestPlugin for scripted experiments. The paper contrasts the LifeSocial.KOM and LibreSocial feature sets, describes the migration to the OSGi/Equinox plugin platform and the HTML5 web interface, and evaluates a network of 64 LibreSocial instances on eight Debian machines (eight slaves plus one master each) under increasing workloads: the total network data rate reaches about 250 MB/s, the average peak is 2,4 MB/s per node, up to 13,000 messages/s in total and about 100 messages/s per node, the hop count stays between 0 and 1 because the routing table holds 3200 entries, and retrieval and storage take about 50 and 110 ms respectively. Storage load is distributed fairly, with a maximum of 4100 unique and 14,000 replicated items per node against averages of about 2500 and 8000, and the cryptographic overhead is small, 42-52 bytes and 1,6-8,1 ms for ECC signing and verification and 3-16 bytes below 1 ms for AES, before the paper closes with future work on security, management and wider deployment.

Publikation (Open-Access-Journalartikel, Special Issue Paper) Overlay (Pastry) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2020 Peer-to-Peer Mechanisms for Fully Decentralized, Secure and Scalable Online Social Networks
Newton Wafula Masinde (Autor); Betreuer Kalman Graffi; Mitautoren der included Papers u. a. Kalman Graffi, Liat Khitman, Iakov Dlikman, Moritz Kanzler, Philipp Mainz, Sebastian Bischoff, Moritz Gericke (HH Düsseldorf, Institut für Informatik)
abstract

This inaugural dissertation at the Mathematisch-Naturwissenschaftliche Fakultät of Heinrich-Heine-Universität Düsseldorf (September 2020, Institute of Computer Science, supervised by Kalman Graffi) develops peer-to-peer mechanisms that make a fully decentralised online social network practical enough to compete with centralised services such as Facebook, whose accumulated scaling costs and single-provider control of data lead to data monetisation and privacy problems. After a comprehensive survey that classifies 23 P2P-based social networks and derives user, system and technical requirements, these are condensed into a four-part architecture of overlay, core framework, social network elements and graphical user interface, and are realised in LibreSocial (formerly LifeSocial.KOM), which has been under continuous development since 2008: a strongly modified FreePastry with a 160-bit identifier space, elliptic-curve keys as node IDs, O(log2 N) routing with Kademlia-like buckets, and PAST extended with UPDATE and DELETE, all built as modular OSGi/Equinox plugins with an HTML5 web interface, which the thesis argues yields a zero-trust network. Systematic benchmarking on a laboratory testbed shows that the social network elements cooperate well and are adequately supported by the underlying P2P services, that the network scales to 2000 nodes without service degradation and handles churn while remaining stable, and that the replication factor (tested with 4, 16 and 32 replicas, derived from the leaf set size) raises memory, storage and bandwidth costs and must grow with the network to keep quality of service. Three mechanisms are then proposed to improve service delivery: metadata-based search over distributed data structures with exhaustive and first-match techniques, the join algorithms Simple LocalJoin, Parallel LocalJoin, asynchronous NetworkJoin and BloomJoin, and binary, deep and customised broad trees, where the right combination of the three choices is decisive for local and network performance; a social caching mechanism that disseminates updates actively to selected friends, with random, trend and social score selection strategies (94,9 %, 93,3 % and 92,4 % cache hit ratio), the social score being preferred, and reaching 99 % cache-based retrievals when combined with the regular cache at the price of extra dissemination traffic; and a capacity management protocol that continuously classifies strong and weak nodes by storage, bandwidth, memory, CPU, battery and connection type and keeps weak nodes out of routing and replica storage, which keeps the network stable even with up to 75 % weak nodes, albeit with some degradation. The thesis concludes that P2P can address both the cost and the privacy concerns of today's social networks and outlines future work on management, larger-scale and realistic Internet deployments.

Dissertation (Inaugural-Dissertation, Heinrich-Heine-Universität Düsseldorf, September 2020) Overlay (Pastry) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2020 Peer-to-Peer based Social Networks: A Comprehensive Survey
Newton Masinde, Kalman Graffi (Technology of Social Networks, Heinrich-Heine-Universität Düsseldorf); arXiv:2001.02611v2 [cs.SI]
abstract

Online social networks, such as Facebook and twitter, are a TABLE I growing phenomenon in today’s world, with various platforms providing S TRUCTURE OF THE S URVEY

Publikation (Survey / Fachartikel) Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2020 Systematic Evaluation of LibreSocial – a Peer-to-Peer Framework for Online Social Networks
Newton Masinde, Liat Khitman, Iakov Dlikman, Kalman Graffi (Technology of Social Networks, Heinrich-Heine-Universität Düsseldorf)
abstract

Peer-to-peer (P2P) networks have been under investigation for several years now, with many novel mechanisms proposed as is evidenced by articles available. Much of the research has focused on showing how the proposed mechanism affects the system performance. In addition, several applications have been proposed to harness the benefits of the P2P networks. Of these applications, online social networks (OSNs) have raised much interest particularly because of the scalability and privacy concerns with centralized OSNs, hence several proposals are in existence. However, accompanying studies on the overall performance of the P2P network under the weight of the OSN applications outside simulations are very few, if any. In this paper, the aim is to undertake a systematic evaluation of the performance of a P2P framework for online social networks called LibreSocial. Benchmark tests were designed, taking into account the random behavior of users, effects of churn on system stability and effect of replication factor. We manage to run benchmark tests for upto to 2000 nodes and show the performance against costs of the system in general. From the results it is evident that LibreSocial provides a performance that is up to par with current user needs.

Dissertation (Preprint-Kapitel,Elsevier, 12.05.2020) Overlay (Pastry) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2020 Systematic Evaluation of LibreSocial — A Peer-to-Peer Framework for Online Social Networks
Newton Masinde, Liat Khitman, Iakov Dlikman, Kalman Graffi (Heinrich Heine Universität Düsseldorf / Honda Research Institute Europe)
abstract

Peer-to-peer (P2P) networks have been under investigation for several years now, with many novel mechanisms proposed as is shown by available articles. Much of the research focused on showing how the proposed mechanism improves system performance. In addition, several applications were proposed to harness the benefits of the P2P networks. Of these applications, online social networks (OSNs) raised much interest particularly because of the scalability and privacy concerns with centralized OSNs, hence several proposals are in existence. However, accompanying studies on the overall performance of the P2P network under the weight of the OSN applications outside simulations are very few, if any. In this paper, the aim is to undertake a systematic evaluation of the performance of a P2P framework for online social networks called LibreSocial. Benchmark tests are designed, taking into account the random behavior of users, effects of churn on system stability and effect of replication factor. We manage to run benchmark tests for up to 2000 nodes and show the performance against costs of the system in general. From the results it is evident that LibreSocial’s performance is capable of meeting the needs of users.

Publikation (Peer-Review-Fachartikel) Overlay (Pastry) Storage & DSS Messaging Monitoring Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2025 Reviving LibreSocial: Error Fixes and a Baseline Evaluation on Modern Systems
Faisal Imam (Betreuung: Prof. Dr.-Ing. Kalman Graffi; Fachbereich 2 Technik, Informatik und Wirtschaft, Technische Hochschule Bingen)
abstract

LibreSocial is a peer-to-peer social network created with vintage Java tools. It is difficult on modern computers to compile, launch, and measure. I re- vive the system on a new host through a binary first approach. I fixed approximately 120 compile and run issues, repaired faulty OSGi metadata, normalized the Java configs, and placed tiny guards so peers launch and monitoring work. Then I created a concise runbook containing the precise configurations and steps necessary to run the setup. While the system was operational, I generated a foundational dataset de- rived from actual executions. Multiple peers participated, established the overlay, and distributed global metrics. I monitored parameters such as node stability, message rates and aggregates, network byte usage, routing hop count, latency for storage and retrieval, along with the impact of the replication factor. The objective is not to document the scale. Rather, the goal is to create a clear and replicable methodology that any future student can utilize to operate and enhance LibreSocial. The result is three contributions. First, a running runtime with trou- bleshooting notes. Second, 20 clear plots and logs out of 900 plots that show the system’s behavior under realistic settings. Third, an artifact pack and instructions that make the experiments reproducible and easy to grow into larger tests or new features.

Masterarbeit Overlay (Pastry) Storage & DSS Messaging Monitoring Search Testing & Evaluation Plugin system App & Kernel PDF · BibTeX
2025 Running LibreSocial: A Legacy Java Project Setup Guide
Faisal Imam (Matrikel-Nr. 2195258, Technische Hochschule Bingen, Fachbereich 2); Gutachter: Prof. Dr.-Ing. Kalman Graffi
abstract

LibreSocial is a fully decentralized peer-to-peer (P2P) framework designed to enable secure, scalable, and privacy-preserving online social networking. This documentation provides a practical guide for running and maintaining the LibreSocial project, which integrates structured P2P overlays, modular plugins, and a zero-trust architecture. It ad- dresses the challenges of dependency resolution, outdated Java APIs, and the migration of legacy code. Key components such as identity management, access-controlled data stor- age, secure communication channels, and distributed data structures (e.g., DHTs, social caches) are emphasized. The documentation also outlines the setup of the development environment, compilation process, and execution flow, while incorporating insights into replication strategies, metadata-based search, and capacity management protocols. Tar- geted at developers and researchers, this document supports the sustainable deployment and extension of LibreSocial in real-world distributed environments. 1 Introduction 1.1 Introduction LibreSocial is a fully decentralized peer-to-peer (P2P) framework that enables secure, privacy-preserving, and scalable online social networking. Originally developed as an academic project and evolved from its predecessor. LibreSocial presents an alternative to conventional centralized social networks by leveraging structured overlay networks and a zero-trust architecture. It incorporates key distributed system components such as identity management, encrypted communication, and modular plugin integration to build a robust social platform free from central control or monetization of user data. This documentation aims to guide users through the process of setting up and running the LibreSocial project, which was originally developed using legacy Java technologies. The project relies on Java 1.6 (compiler compliance level) and JavaSE-1.7 runtime environ- ment, making it incompatible with modern IDEs (eg. Eclipse 2023 or JetBrai

Dokumentation / HowTo (Setup-Guide im Rahmen einer Masterarbeit, Endfassung) Overlay (Pastry) Storage & DSS Security Plugin system App & Kernel PDF · BibTeX
2025 Running LibreSocial: A Legacy Java Project Setup Guide (Entwurfsversion TH Bingen)
Faisal Imam (Matrikel-Nr. 2195258); Gutachter: Prof. Dr.-Ing. Kalman Graffi; Technische Hochschule Bingen, Fachbereich 2
abstract

LibreSocial is a fully decentralized peer-to-peer (P2P) framework designed to enable secure, scalable, and privacy-preserving online social networking. This documentation provides a practical guide for running and maintaining the LibreSocial project, which integrates structured P2P overlays, modular plugins, and a zero-trust architecture. It ad- dresses the challenges of dependency resolution, outdated Java APIs, and the migration of legacy code. Key components such as identity management, access-controlled data stor- age, secure communication channels, and distributed data structures (e.g., DHTs, social caches) are emphasized. The documentation also outlines the setup of the development environment, compilation process, and execution flow, while incorporating insights into replication strategies, metadata-based search, and capacity management protocols. Tar- geted at developers and researchers, this document supports the sustainable deployment and extension of LibreSocial in real-world distributed environments. 1 Introduction 1.1 Introduction LibreSocial is a fully decentralized peer-to-peer (P2P) framework that enables secure, privacy-preserving, and scalable online social networking. Originally developed as an academic project and evolved from its predecessor. LibreSocial presents an alternative to conventional centralized social networks by leveraging structured overlay networks and a zero-trust architecture. It incorporates key distributed system components such as identity management, encrypted communication, and modular plugin integration to build a robust social platform free from central control or monetization of user data. This documentation aims to guide users through the process of setting up and running the LibreSocial project, which was originally developed using legacy Java technologies. The project relies on Java 1.6 (compiler compliance level) and JavaSE-1.7 runtime environ- ment, making it incompatible with modern IDEs (eg. Eclipse 2023 or JetBrai

Studienarbeit/Masterarbeit (Entwurfs- bzw. Rohfassung mit Platzhaltern) Overlay (Pastry) Storage & DSS Security Plugin system App & Kernel PDF · BibTeX