Application Signature: An Analytical Technique to Model System Performance

Advances in digital computers have been spectacular but increasingly complex to model. Even the cycle-accurate simulators that are costly to develop and run have questionable accu-racy. This paper provides a simple, fast but accurate, scientifically proven analytic technique to model accurately the performance of real applications. The method creates two profiles as a function of time or problem sizes. The first profile, Hardware Signature, which reveals com-puter hardware characteristics, is obtained by running a universal benchmark, HINT (Hierar-chical INTegration) or by running an analytical model, AHINT. The second profile, Application Signature (APPMAP), which divulges intrinsic application requirements, can be obtained by simple methods as outlined in the paper. The proposed model convolutes the hardware signature of a system under study and the application signature of a given application to predict accurately the application’s performance. The model was used to predict SPEC benchmarks. The model was tested using 25,000 per-formance measurements and validated using standard statistical methods. such as Pearson’s correlation, Spearman’s rank correlation and maximum error in linear prediction. Furthermore, for a given application signature, the input parameters of the analytical hardware model can be 1 changed to study the important what-if scenarios such as the performance impact of changing the size, bandwidth or latency of memory and cache hierarchy; numerical precision and clock rate of the processor. We have recently achieved an application performance capability that we believe is unprecedented in its accuracy.