Parallel Computing
Ranking of various HPC architectures has been of keen interest to many research centers and application designers. With the introduction of multi-core processors, clusters, and grids, many parallel computer configurations emerge. Hence, given a compute intensive application, determining an efficient configuration is challenging. Highly centralized systems are obviously faster than distributed one; however there is a trade-off with cost. Moreover, for a highly parallel application, a lower cost distributed system could yield better performance than a centralized configuration.
In this research, we propose a methodology for deciding which computation configuration is 'better' for a given compute intensive application. The term 'better' takes into consideration performance and cost. Such models can be used by computing centers in system acquisitions and by application scientists to improve the performance of their HPC applications.
The two main configuration models considered here are: the centralized and the distributed (server, cluster, and grid). We envision a bipartite model comprising of:
- An application performance analyzer and parameter extractor.
- A configuration selector.
- An optimizer.
We will use a wide variety of computation-intensive applications as our workloads for testing our tool.
Key Investigators
Dr. Mohamed Saleh, Assistant Professor, Faculty of Engineering
Dr. Mohamed Zahran, Assistant Professor, City University of New York
Rehab ElKady
Back
|
