Wide area distributed storage systems leverage the scale of the Internet to provide vast amounts of data storage capabilities. The typical building block is a peer-to-peer system that can pool disparate distributed resources into a cohesive storage system. However, the participant nodes are relatively unreliable and thus, distributed storage systems must take care to provide replicas of data objects. A common problem is how to allocate objects to nodes in the system and previous work has

more »

... ed objects based on popularity. However, these schemes are not general enough to minimize latency or deal with redundancy schemes such as erasure codes which are based on replication of object fragments. In this paper, we present a strategy for placement of replica fragments particularly with respect to the minimization of data communication latencies.