Squirrel: A decentralized peer-to-peer web cache

Sitaram Iyer, Antony Rowstron, Peter Druschel
PODC '02: Proceedings of the twenty-first annual symposium on Principles of distributed computing

Overview

Squirrel is based on Pastry. Using Pastry Squirrel identifies the home node for an object. Squirrel has two variants, the home store scheme and the directory scheme. With the home store scheme a home node is responsible for having a copy of the objects that are mapped to it. If the home node has a fresh copy then it sends the copy to the requester. If the home node has not the object cached then it will first retrieve the object from the origin server and subsequently send it to the requester.

With the directory scheme the home node maintains a table with pointers to node that have recently requested the object and are therefore likely to have a cached copy. When a home node receives a request it forwards the request to a randomly chosen node from the directory.

Both schemes achieve comparable performance in terms of hit ratios and latency. The directory scheme does better load balancing. Directories of popular object changes rapidly and the load for these objects is spread over many nodes. In practice the home store scheme achieves a lower load.