Contribution incentives for hybrid peer-to-peer systems

Peer-to-peer protocols have become an important building block for constructing scalable Internet applications. Hybrid peer-to-peer protocols in particular, which utilize a centralized control server to ease management and facilitate node bootstrapping, have been shown to be very effective in Internet content distribution. However, the viability of peer-to-peer systems hinges on the willingness of nodes to contribute their resources. As a result, incentive mechanisms that encourage such contributions are instrumental to these systems' continued success.;This thesis examines how contribution incentives for hybrid peer-to-peer systems should be designed. We first investigate the properties of BitTorrent, arguably the most popular hybrid peer-to-peer protocol, whose success might be explained in part by its built-in contribution incentive scheme. To that end, we run extensive experiments on both a controlled testbed and in real-world BitTorrent swarms. Our results showcase the efficiency of BitTorrent's incentives and demonstrate that the protocol discourages free-riding by rewarding contributors---peers who upload to others---with short download completion times.;Additionally, in an effort to better understand incentive limitations, we study the behavior of selfish peers, who abuse protocol mechanisms to maximize their download throughput while minimizing their own uploads to others. We find the protocol quite robust against such peers, in that it prevents them from degrading the quality of service for compliant nodes. Based on our results, we identify underlying principles that enable this robustness.;Having investigated how BitTorrent's incentives successfully facilitate content distribution, we seek to extend our results to a different application scenario that requires more complex incentives---peer-to-peer data storage for Web applications. We design and implement Cloudfarm, a hybrid peer-to-peer storage substrate that supports the participation of end-user machines in the storage of Web content. Cloudfarm enables Web clients to persistently store their data on other clients, thereby disassociating user-generated content from the application providers who originally controlled access to it. This in turn facilitates more flexible data sharing across applications.;Cloudfarm's incentive requirements are fundamentally different from those of BitTorrent. As a result, Cloudfarm incorporates a novel contribution incentive scheme that spans several metric dimensions, including peer round-trip latency and bandwidth service. We port an existing open-source photo gallery application to the Cloudfarm architecture and show experimentally that it incurs acceptable webpage load overhead. We also validate the effectiveness of the incentive scheme to encourage peer contributions.;Our results not only explain the impact of existing incentive mechanisms in hybrid peer-to-peer systems, but also extend their applicability to distributed data storage scenarios.