AbstractsEngineering

PeerSelector: A framework for grouping peers in a P2P system

by Rakesh Kumar




Institution: KTH Royal Institute of Technology
Department:
Year: 2012
Keywords: Engineering and Technology; Teknik och teknologier; Technology; teknik; Master of Science -Communication Systems; Teknologie masterexamen - Kommunikationssystem
Record ID: 1336115
Full text PDF: http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-99561


Abstract

This master thesis presents a framework called PeerSelector that has been designed and implemented to group peers in a Peer-to-Peer (P2P) system according to certain criteria. The framework is portable and can be deployed with any distributed P2P system. We devised the framework with such functionality in mind because we consider that grouping peers according to certain criteria can benefit the users of the system by providing them with more flexibility to group peers according to their own interests, without depending on entities such as ISPs for peer clustering. We designed and implemented a modular architecture for the framework. More specifically, PeerSelector consists of modules that implement basic functionalities such as grouping peers according to geo-location, RTT-based latency, and the number of AS (Autonomous System) hops. When peers are grouped according to the respective metrics, they are stored in queues, namely the distance, latency, and hop-count queues. Any P2P system that is integrated with our framework fetches peers from such queues, on demand. The results from the framework functionality testing show that the framework is successfully able to cluster peers according to the user's indicated interest. In addition, the framework has been integrated with two existing P2P protocols with minor adjustments, confirming the flexibility and portability of the framework across applications. We have carried out experiments to investigate if using our peer clustering techniques helps a P2P client increase its download performance. In our experiments with a live swarm, we learned that grouping peers according to geo-location does not influence the download performance drastically: download performance increases slightly or remains the same for almost 75 percent of the cases. For the two other clustering metrics, latency and AS hops, our preliminary experimental results don't always show an improvement of the client's download performance. ; Denna examensarbete presenterar en konstruktion kallad PeerSelector som har blivit designat och implementerat för att gruppera klienter i ett P2P-system efter vissa kriterier. Konstruktionen är portabel och kan användas med alla distribuerade P2P-system. PeerSelector är utformat så att användarna kan gruppera klienterna efter deras egna intressen, vilket ger ökad flexibilitet, utan att vara beroende av enheter så som ISP. Vi har designat och använt en modulär arkitektur för konstruktionen av PeerSelector. Detta betyder att den består av moduler som utför basfunktioner så som att gruppera klienterna efter geografiskt läge, RTT-baserad fördröjning, och antal AS-stopp. När klienterna är grupperade efter respektive ämnesdomän, är de förpassade in i olika kösystem, i synnerhet efter sträcka, fördröjning och stopp-räkning. Alla P2P-system som är integrerade med vår konstruktion hämtar på begäran klienter från dessa kösystem. Resultaten från funktionalitetstester visar att PeerSelector kan gruppera klienter efter användarens intressen. Vidare…