Adaptive placement of parallel Java agents in a scalable computing cluster

This paper describes a framework for parallel computing in a locally confined, scalable computing cluster (SCC) using Java agents. The framework consists of a programming model with agents and asynchronous invocations, and a scheme for adaptive placement of multiple agents in an SCC. Our scheme is geared to improve the overall performance by a dynamic match between the available cluster resources and the execution requirements of the agents. This is accomplished by agent migrations among the nodes using on-line algorithms for load leveling and reduction of the interagent communication overhead. Simulations of several examples show that our scheme outperforms a static agent placement scheme by as much as 40% for the test cases.