Pro-active performance management of distributed applications

Abstract

Self-managing systems able to dynamically re-configure with respect to time-varying workload mixes and changing system resource availability are of growing importance in heterogeneous multicomputer environments. Advanced performance evaluation techniques are needed to induce and assess the impact of such re-configurations where guaranteeing timeliness of reconfiguration activities is particularly challenging. A whole new class of methods supporting "pro-active" adaptivity based on the predicted system state at the re-configuration instant are needed to alleviate the shortcomings of "re-active" systems which bring reconfigurations in effect after the system state has changed. In this paper we argue for self-contained performance management of distributed applications, i.e. integrated performance tuning with the ability to automatically adapt the application behavior to the performance characteristics of the execution environment. Particularly, we study pro-active performance management for distributed simulation based on the Time Warp protocol executing on a network of workstations (NOWs).