Resource Localization Using Peer-To-Peer Technology for Network Enabled Server.

Abstract

(eng) DIET (Distributed Interactive Engineering Toolbox) is a set of hierarchical components to design Network Enabled Server systems. These systems are built upon servers managed through distributed scheduling agents for a better scalability. Clients ask to these scheduling components to find servers available (using some performance metrics and information about the location of data already on the network). Our target architecture is the grid which is highly heterogeneous and dynamic. Clients, servers, and schedulers are better connected in a dynamic (or peer-to-peer) fashion. One critical issue to be solved is the localization of resources on the grid. In this paper, we present the use of an asynchronous version of the Propagate Information with Feedback algorithm to discover computation resources in Network Enabled Servers with distributed schedulers and arbitrary networks. Resource discovery uses peer-to-peer connections between components. Our implementation is based on JXTA from Sun Microsystems and DIET developed in the GRAAL team from INRIA. The algorithm and its implementation are discussed and performance results to show the benefit of this approach are given from experiments over the VTHD network which connects several supercomputers in different research institutes through a high-speed network.