Performance Forecasting : Characterization of Applications on Current and Future Architectures

A common approach to studying future computer systems is to build simulators that accurately model the behavior of applications and architectures. While this approach can lead to quantitative information about the systems performance it has one serious drawback: simulating large applications with large data sets can be prohibitively time-consuming. In this paper we propose an alternative approach. We model the performance of an application with \Resource Usage Equations," verify these equations with acual program runs on current systems and current data sets, and then extrapolate the system behavior, scaling up architectural components and data set parameters.