Application placement using performance surfaces

Abstract

Heterogeneous parallel clusters of workstations are being used to solve many important computational problems. Scheduling parallel applications on the best collection of machines in a heterogeneous computing environment is a complex problem. Performance prediction is vital to good application performance in this environment since utilization of an ill-suited machine can slow the computation down significantly. The heterogeneity of the different pieces composing the parallel platform (network links, CPU, memory, and OS) makes it incredibly difficult to accurately predict performance. This paper addresses the problem of network performance prediction. Since communication speed is often the bottleneck for parallel application perfomance, network performance prediction is important to the overall performance prediction problem. A new methodology for characterizing network links and application's need for network resources is developed which makes use of performance surfaces (Clement et al., 1998). Mathematical operations on the performance surfaces are introduced that calculate an application's affinity for a network configuration. These affinity measures can be used for the scheduling of parallel applications.