الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 174 |  |  |
| | | from | 174 | | |
Abstract
Performance evaluation of parallel processing computer systems has been of an increasing demand to both
speciailists and researchers.
Despite of the rich literature in the field of performance modeling, there has been always request for truthful characterization of applied benchmarks as well as more flexible modeling along well defined evaluation criterion.
So many factors and variations need to be considered when evaluating the effect of such variations on the overall system performance. The target is to identify a subjective selection criterion to enable architects of tuning their architectural configurations to the processing needs of the applied workloads.
Consequently , many different but highly interacting factors need to be investigated including processor speed and number of processors, memory modules, buses as well as proper scheduling of workload over hardware resources.
Throughout this research, an in-depth study has been conducted to cover different performance modeling techniques. Main focus has been given to the problem of simulation- modeling as well as the performance metrics identification.
A complete evaluation model for parallel processing computers has been developed.
This model is based on three attribute simulators, the architecture topology simulator, the benchmark characterization model and the performance criterion.
A number of proposed performance figures has been introduced to make the analysis more subjective from the perspective of cost effectiveness. These include utilization, speed-up factor, memory and bus traffic threshold and cost of gained performance factor.
The model has successfully simulated a number of bus oriented parallel computer .
Results have been validated via comparing to analytical results of typical combinatorial searching workloads.
The analysis of the presented results have proven to give truthful judgment of cost effective enhanced features in terms of performance.
The results have favored the enhanced memory and process acheduling selection over processors in moderate computational-bound-type problems.