Early prediction of MPP performance: The SP2, T3D, and Paragon experiences

Zhiwei Xu, Kai Hwang
1996 Parallel Computing  
The performance of Massively Parallel Processors (MPPs) is attributed to a large number of machine and program factors. Software development for MPP applications is often very costly. The high cost is partially caused by lack of early prediction of MPP performance. The program development cycle may iterate many times before achieving the desired performance level. In this paper, we present an early prediction scheme for reducing the cost of application software development. Using workload
more » ... is and overhead estimation, our scheme optimizes the design of parallel algorithm before entering the tedious coding, debugging, and testing cycle of the applications. The scheme is applied at user/ programmer level, not tied to any particular machine platform or to any specific software environment. We have tested the effectiveness of this early performance prediction scheme by running the MIT/STAP benchmark programs on a 400-node IBM SP2 system at the Maui High-Performance Computing Centre (MHPCC), on a 400-node Intel Paragon system at the San Diego Supercomputing Centre (SDSC), and on a 128-node Cray T3D at the Cray Research Eagan Centre in Wisconsin. Our prediction is shown rather accurate compared with the actual performance measured on these machines. We use the SP2 data to illustrate the early prediction scheme. We provide a systematic procedure to estimate the computational workload, to determine the application attributes, and to reveal the communication overhead in using MPPs. These results can be applied to develop any MPP applications other than STAP radar signal processing, from which this prediction scheme was developed.
doi:10.1016/0167-8191(96)00034-8 fatcat:e2hwjnp5efdvjjsaoiidhr3qei