Predicting Performance of Parallel Applications Based on Distributed Shared Memory System

摘要

We proposed a method to evaluate the performance of applications running on a distributed shared memory system. In the method, we established communication models in application level, DSM system level and shared memory consistency protocol level, respectively. Then we fusing these three models into an integrated predicting model. We analyzed LU algorithm and evaluated its performance. The experimental results show that this method is available, and can observe the performance effects come from the changing of system parameters. and can be applied to optimize parallel algorithm and compare alternative system designing blueprint.