Accurate Performance Modelling and Prediction of Cluster Computers

A key issue facing academic and industrial organizations with high performance computing requirements is which cluster computer configuration will be optimal for their applications. The diverse and expanding range of such configurations means that only small prototypes are likely be available for evaluation. This project will address this issue by developing efficient tools and methodologies to accurately characterise and model both real-world applications and (small-scale) cluster configurations. Of key importance will be to capture secondary effects such as contention and the degree of CPU involvement in communication. These tools will be used to accurately predict the application's performance on medium to large-scale systems