Scalable Data Sharing Without Centralized Trust

Data sharing on the Internet increasingly relies on centralized trust. Popular services mediate data exchange among millions of people by building upon trusted, large-scale infrastructure, e.g., data centers and content distribution networks. While scalable and robust, this approach has drawbacks: the infrastructure is costly and controlled by a single organization, eroding privacy and making censorship trivial.;Peer-to-peer (P2P) systems provide a promising alternative for scalable data sharing without relying on centralized trust. Rather than using dedicated infrastructure resources, P2P designs achieve scalability by aggregating the collective resources of participants. P2P scalability depends on participants contributing their resources as they make demands of others. But, without any reason to do so, many users simply "free ride"; i.e., consuming resources without contributing. Today, BitTorrent is one of the most popular P2P applications, in part because it was designed with contribution incentives explicitly intended to discourage free riding and strongly link each client's performance with their level of contribution.;In this dissertation, I show that BitTorrent's incentive scheme is largely ineffective in practice, resulting in poor performance and availability. In its place, I describe two new protocols to strengthen P2P contribution incentives: One Hop Reputations and Contracts. I find that BitTorrent's incentives can be circumvented by a strategic client, in part because relationships between participants are short-lived. One Hop Reputations strengthens incentives by discovering long-term, indirect relationships, allowing would-be trading partners to be evaluated using richer history and information. Contracts adapts this approach to P2P live video streaming. Since live streaming systems cannot reward users with increased download rates. Contracts uses one hop propagation of contribution records to detect and reward globally beneficial contributions with improved robustness.;To evaluate One Hop Reputations and Contracts. I have applied each to a representative P2P design. When applied to BitTorrent, One Hop Reputations improves download performance and provides high coverage of possible trading partners. Moreover, the protocol achieves these benefits with modest overhead, retaining scalability. When applied to the popular PPLive video streaming service, Contracts increases the fraction of clients experiencing loss-free playback relative to the unmodified implementation and strengthens incentives by linking each user's playback quality to their level of contribution. Taken together, One Hop Reputations and Contracts significantly improve performance and incentives for P2P data sharing systems, providing an alternative to services dependent on centralized trust.