TU Berlin

Internet Network ArchitecturesPublications by Type: PhD Theses

Page Content

to Navigation

Publications by Type: PhD Theses

Dynamics and Cooperation: Algorithmic Challenges in Peer-to-Peer Computing
Citation key S-DCACPC-08b
Author Schmid, Stefan
Year 2008
Month April
Note Diss. ETH No. 17731, TIK-Schriftenreihe Nr. 96
School ETH Zurich
Abstract Peer-to-peer (p2p) computing is one of the most intriguing new networking paradigms of the last ten years, and many applications today use peer-to-peer technology, e.g. for large distributed computations, for file sharing, or for live media streaming. At the heart of the paradigm lies the idea of leveraging the resources of the system's participants. Thus, potentially scalable and robust architectures can be built. However, making use of the decentralized resources is challenging. The peers are under the control of the individual users who may only connect to the network for a short period of time. Consequently, there are frequent membership changes and p2p systems are highly dynamic. In addition to regular joins and leaves, the participating machines (often unreliable desktops) may crash. Peer-to-peer solutions are also faced with the fact that it is not always in the (anonymous) users' interest to contribute their resources. Rather, a user may be selfish and seek to exploit the system without reciprocating. This dissertation studies the challenges of the dynamics in p2p computing and of cooperation. We describe a system which is based on a hypercubic topology and which applies algorithms that maintain desirable network properties despite worst-case joins and leaves; these algorithms can also be used for pancake graphs. Besides membership dynamics, we investigate dynamic changes of the available bandwidth between two peers, and we analyze the throughput of different transfer protocols. In order to emphasize the importance of the cooperation challenge, we conduct a case study of BitTorrent–one of the most traffic intensive applications on the Internet–, and show that today's peer-to-peer networks still fail to fend off uncooperative peers. A game-theoretic analysis of a p2p network creation game is presented which estimates the impact of selfish behavior. We find that both the performance and the stability of a system can suffer severely. In addition, this dissertation introduces a mathematical framework which allows us to evaluate a system's robustness to malicious attacks; the framework is also useful for the analysis of social networks. The theoretic findings are complemented by a case study which identifies vulnerabilities in the popular Kad network.
Link to publication Download Bibtex entry


Quick Access

Schnellnavigation zur Seite über Nummerneingabe