A scalable simulation framework for evaluating thermal management techniques and the lifetime reliability of multithreaded multicore systems

Abstract

It has become increasingly challenging to understand supercomputers behavior and performance as they grow. New hurdles in scalability, programmability, power consumption, reliability, cost, and cooling are emerging. This paper introduces the integrated power, area, temperature, reliability modeling framework in the open, modular, multiscale, parallel Structural Simulation Toolkit (SST) to help evaluate new technologies and guide design of future computers. In this study, the simulation framework is used to evaluate different dynamic thermal management techniques, in terms of power, temperature and reliability, on multicore systems running multithreaded and more irregular applications. Simulation results shed some new light on application-aware, performance/power efficient thermal and reliability management policies of multithreaded multicore systems.