Reperasure: Replication Protocol using Erasure-code

Peer-to-peer overlay networks offer a convenient way to host an infrastructure that can scale to the size of the Internet and yet stay manageable. These overlays, which are essentially self-organizing distributed hash tables (DHT), however, face challenges of data reliability because of the dynamic nature of the system. Furthermore, in order to see wider adoption, it is time to design support for generic replication mechanisms capable of handling arbitrary update requests – most of the existing proposals are deep archival systems in nature. Utilizing the fact that DHT can function as a super-reliable and high performance disk when data stored inside are erasure coded, we believe practical and simple protocols can be designed. In this paper, we introduce the reperasure protocol, a layer on top of the basic DHT, which efficiently supports strong consistency semantic with high availability guarantee. By relieving the DHT layer out of replication duty inside, a cleaner overall architecture can be derived because of the clear division of responsibility.