A Cost-Benefit Approach to Resource Allocation in Scalable Metacomputers

R. Sean Borgstrom, Johns Hopkins University

The performance of a network of machines improves if they can share computing resources; that is, if a job arriving at one machine can be executed on another. To achieve the best performance, the scheduler that decides which machine executes each incoming job must use an efficient and intelligent strategy. This work proposes three new strategies for job assignment with beneficial theoretical properties, experimentally proven to perform well in practice. It also includes a complete system, the Jini CBF Package, used to perform intelligent job assignment using one of these strategies on a Jini network.