Performance Modeling of SIMD and MIMD Parallel Computers using Neural Networks

点击翻译按钮获取中文摘要

The architecture of future teraflop computers is under intense study in universities, government and industrial laboratories. The objective of this research is to model parallel applications on some of the recently available commercial architectures. The goal is to develop performance modeling methods that will scale to future machines. Performance evaluation of parallel computer systems and applications running on these systems is a complex task because of the large number of execution parameters involved. This project investigates the use of robust modeling techniques based on neural networks to analyze the data and predict performance on current as well as future machines. This analysis will also be useful in providing information about the relevant measurements to be stored for useful prediction.