Computing with data non-determinism: Wait time management for peer-to-peer systems

Abstract

One of the unusual challenges faced by peer-to-peer algorithms as opposed to classical distributed algorithms is that these have to compute with data non-determinism where there is no guarantee that data from a particular node will be delivered in time or delivered at all. Each distributed component not only has to work in such environment, but also has to decide how long to wait for peers to respond. A large wait can inordinately delay entire computation. A smaller wait can reduce the coverage of the algorithm and drastically degenerate the quality of final result produced. In this paper, we investigate a set of wait management schemas which tries to minimize the overall completion time and maximize the quality of solution. We present detail analyses of the schemas for various scenarios including large power law networks and forms of non-responsiveness. The work promises dramatic improvement in peer-to-peer search and other computations. © 2007 Elsevier B.V. All rights reserved.