| Abstract
|
In this paper, we examine the performance … In this paper, we examine the performance of a parallel application implemented in both shared variable and message passing styles. Our purpose is to illuminate the differences between the programming models and show how these differences affect the performance of the programs when executed on systems incorporating hundreds of processing elements. First, we present the programming models used for the implementations. Then we examine the costs associated with each model, focusing on interprocessor communication and synchronization. Strategies for minimizing data communications costs are discussed and confirmed through simulation. Also, architectural features are identified which have a substantial impact on shared variable and message passing performance. variable and message passing performance.
|
