Better than 1 Hop Lookup Performance with Proactive Caching

Abstract

High lookup latencies prohibit peer-to-peer overlays from being used in many performance intensive applications, even though they provide self-organization, scalability, and failure resilience. In this paper, we show that lookup performance of structured DHTs can be improved to any desirable constant, even under 1 hop, by controlled proacive replication. By exploiting the popularity distribution of objects, we can minimize the number of replicas and reduce the storage and bandwidth cost of replication. This enables structured DHTs to efficiently support a wide variety of latency sensitive applications. We describe three different applications, namely DNS, web access, and content distribution, and show how they can derive significant performance gains by using DHTs.