Scalability analysis of parallel system based on memory efficiency on distributed memory architecture

07/16/2010 00:00:00

---

 

 

 

 

The increasing demand of greater computational speed has motivated scientist to make use of parallel computations. Nevertheless, the performance of parallel system is not straight forward; it is affected in fact by many factors. One significant factor that limits the performance gain is the inefficiency of memory used by parallel systems. Memory efficiency describes how well a parallel program is designed to exploit the available memory in parallel computers, particularly in distributed systems, and whether a parallel program is scalable in its memory usage. Two problems are addressed in this article, i.e matrix-matrix multiplication, and matrix-vector multiplication. It is observed that data replication that is frequently adpted in a parallel program in order to reduce communication overhead causes bad utilization of available meory, and hence a degradation of the scalability of the parallel system.

Keywords : memory efficiency, parallel computation, performance, scalability.