Automatic Reconfiguration for Maximizing System Performance on Software Distributed Shared Memory Systems

Y.-C. Zhuang, C.-K. Shieh, T.-Y. Liang, and J.-C. Ueng (Taiwan)


Performance Prediction, Distributed Shared Memory


This paper presents an automatic reconfiguration runtime system for predicting and tuning the performance of iterative parallel programs on distributed shared memory (DSM) systems. We address the problem of maximizing application speedup through performance prediction to automatically select an appropriate number of processors to run. The basic concept of this mechanism is to predict the program performance based on the runtime information of the parallel programs without any user intervention. With the support of this performance prediction mechanism, DSM systems can adjust system configuration to the appropriate set of workstations for approaching the peak performance. Experiments show that applying the performance prediction policy is necessary for maximization the utilization of the computing resources and program speedup in DSM.

Important Links:

Go Back