Title: Processor Modeling using Monte Carlo Techniques: An Alternative to Detailed Simulation

Speaker: Jeanine Cook, New Mexico State University

Date/Time: Thursday, April 8, 2010, 11:00 a.m.        

Location: CSRI Building/Room 279 (Sandia NM)

Brief Abstract: Cycle-accurate simulation of contemporary, multi-core processors is difficult for two primary reasons: the lack of publicly available simulators and the large wall-clock time associated with simulating these architectures. We have developed a statistical technique based on Monte Carlo methods to model processor performance and behavior.   Models are developed using processor characteristics obtained from manuals and from targeted microbenchmarking. Application characteristics and metrics collected from application execution are used as parameters to the models. The methodology has been applied to single- and multi-core architectures, and to both in-order execution and out-of-order execution processors. The most recently developed models and their respective results will be presented.

CSRI POC: Arun Rodrigues, (505) 284-6090