An automatic trace based performance evaluation model building for parallel distributed systems

Ahmad Mizan, Greg Franks. An automatic trace based performance evaluation model building for parallel distributed systems. In Proceeding of the second joint WOSP/SIPEW international conference on Performance engineering. 2011. [doi]

Abstract

Performance models can be built at early stages of software development cycles to aid software designers to assess design alternatives and identify fundamental design problems before the implementation cycles starts. These models are flexible for varying operational conditions and design alternatives; however, their creation is not trivial and requires considerable effort. An automatic performance model making system has been developed which uses traces of events from an instrumented executable program in a distributed computing system to create layered queuing network (LQN) performance evaluation models. The model construction process is independent of local time of the geographically dispersed nodes by using a new format for logical time. Low communications overhead and simplified interaction identification are the main characteristics of this format. The system can identify Synchronous, Asynchronous, Forward and Loop interaction types between the nodes in which tasks have internal concurrency.