Load sharing for optimistic parallel simulations on multi core machines
ACM SIGMETRICS Performance Evaluation Review
| Hi-index | 0.00 |
In this paper we show how program code and performance models can be made to cooperateseamlessly to support continuous software performance testing throughout thedevelopment lifecycle. We achieve this by extending our existing VEXtool for executing programs in virtual time so that events that occurduring normal execution and those that occur during the simulation of a performance model can bescheduled on a single global virtual time line. The execution time of anincomplete component of an application is thus estimated by a performance model, whilstthat of existing code is measured by instrumentation that is added dynamicallyat program load time. A key challenge is to be able to map some or all of the resourcesin a performance model to the real resources of the host platform on which theapplication is running. We outline a continuous performance engineering methodologythat exploits our unified framework and illustrate the principles involved byway of a simple Java application development case study.