Distributed File Structures in a Peer-to-Peer Environment

As computing becomes increasingly distributed, the need for accessing and manipulating distributed data becomes prominent. The focus so far has been on developing fault-tolerant, load-balanced and scalable techniques for distributing data across the network without placing much importance on the semantics of the applications built on top of them - much in the way a centralized file system simply serves disk pages without handling their semantics. In this paper, we continue this "black-box" approach and extend it to the next level. We take standard file structures and distribute them across a peer-to-peer network and test their performance. We then identify performance bottlenecks and propose ways of overcoming them. These results, we believe, exhibit the potential of seamlessly extending computing applications assuming standard (centralized) file I/O capabilities to work in distributed environments.