Distributed hash tables are increasingly being proposed as the core substrate for content delivery applications in the Internet, such as cooperative Web caches, Web index and search, and content delivery systems. The performance of these applications built on DHTs fundamentally depends on the effectiveness of request routing within the DHT. In this paper, we show how to use soft state to achieve routing performance that approaches the aggressive performance of one-hop schemes, but with an order of magnitude less overhead on average. We use three kinds of hint caches to improve routing latency: local hint caches, path hint caches, and global hint caches. Local hint caches use large successor lists to short cut final hops. Path hint caches store a moderate number of effective route entries gathered while performing lookups for other nodes. And global hint caches store direct routes to peers distributed across the ID space. Based upon our simulation results, we find that the combination of the hint caches significantly improves Chord routing performance: in a network of 4,096 peers, the hint caches enable Chord to route requests with average latencies only 6 % more than algorithms that use complete routing tables with significantly less overhead. 1 1

A major hurdle to deploying a distributed storage infrastructure in peer-to-peer systems is storing data reliably using nodes that have little incentive to remain in the system. We argue that a node should choose its neighbors (the nodes with which it shares resources) based on existing social relationships instead of randomly. This approach provides incentives for nodes to cooperate and results in a more stable system which, in turn, reduces the cost of maintaining data. The cost of this approach is decreased flexibility and storage utilization. We describe our approach and sketch two applications for which this approach is viable: a cooperative backup system and a Usenet replacement.

Recent research has convincingly demonstrated that one can build scalable, robust, and efficient distributed hash tables (DHTs). However, most applications built over DHTs either have very simple application requirements, in which case strictly layering them over a generic DHT is straightforward, or use specially modified DHT implementations, which prevents the use of off-the-shelf DHT software or third-party DHT services. An unresolved issue is whether applications with more sophisticated requirements can be feasibly implemented over a generic DHT. This question is probably best answered by building and evaluating DHT-based applications. To this end, this paper reports on our (somewhat mixed) experiences with building Place Lab, an end-user positioning system for locationenhanced applications, on top of a generic DHT. While certainly not the first foray into DHT applications, Place Lab differs from most prior efforts in that its application interface, based on geographic range queries, is far more demanding than the typical DHT exact-match lookup. Moreover, operational concerns led to a “layered ” system architecture in which the deployment and management of the DHT is completely decoupled from Place Lab’s other system components. In this paper we present a design and our experiences with a prototype implementation built entirely on top of a generic DHT. 1

Distributed hash tables (DHTs) have been proposed as a generic, robust storage infrastructure for simplifying the construction of large-scale, wide-area applications. For example, UsenetDHT is a new design for Usenet News developed in this thesis that uses a DHT to cooperatively deliver Usenet articles: the DHT allows a set of N hosts to share storage of Usenet articles, reducing their combined storage requirements by a factor of O(N). Usenet generates a continuous stream of writes that exceeds 1 Tbyte/day in volume, comprising over ten million writes. Supporting this and the associated read workload requires a DHT engineered for durability and efficiency. Recovering from network and machine failures efficiently poses a challenge for DHT replication maintenance algorithms that provide durability. To avoid losing the last replica, replica maintenance must create additional replicas when failures are detected. However,