Evaluation of SimPoint for Specific Architectural Studies

Abstract

As microprocessor complexity increases, the time required to complete a thorough simulation of an entire benchmark will become a bigger and bigger challenge that computer architects have to face. This simulation time is orders of magnitude greater than the time required to run the benchmark on real hardware, thus, it is no longer feasible to simulate long benchmarks in their entirety. Given such a scenario, there is a definite need to reduce this simulation time so that computer architecture research can be performed efficiently in the future. SimPoint [1] is a popular tool used to select representative samples of a benchmark. These samples are then used for simulation, instead of the entire benchmark, resulting in a significantly shorter simulation time. Since these samples will be used by computer architects to determine performance improvements, it is imperative that there should be little compromise in accuracy. Otherwise, the results obtained when using SimPoint could not be trusted. The goal of our project is to evaluate the effectiveness of using SimPoint in determining the relative performance improvement of a particular architectural enhancement. The results will either validate or invalidate the effectiveness of using SimPoint for particular architecture studies.

Extracted Key Phrases

Cite this paper

@inproceedings{NarasimanEvaluationOS, title={Evaluation of SimPoint for Specific Architectural Studies}, author={Veynu Narasiman and Aater Suleman} }