Dynamic Power and Thermal Monitoring and Management for High-Performance Microprocessors

Jun Yang, ECE/PITT

Tuesday, December 5, 2006
Noon - SENSQ 5317
Free pizza for attendees starting at 11:45 a.m.

Advisor: Rami Melhem

Abstract

The evolution of microprocessors has been hindered by their increasing power consumption and the speed heat is generated on-die. High temperature impairs the processor reliability and reduces its lifetime. While hardware level dynamic thermal management (DTM) techniques can effectively lower the chip temperature when it surpasses certain threshold, they inevitably come at the cost of performance downgrading such as voltage and frequency scaling. To ensure an effective control of the chip temperature, it is imperative to be able to monitor the temperature variations across the die timely and accurately. Most current techniques rely on on-chip thermal sensors, typically one or two, to report the temperature of the processor. Unfortunately, the significant variation in chip temperature both spatially and temporally exposes the limitation of the sensors since hot spots migrate with workloads. We present an alternative approach to tracking chip temperature through an OS resident software module that generates live power and thermal spectral distributions of the processor. We developed such a software thermal sensor (STS) with low overhead in a Linux system with a Pentium 4 Northwood core. The software thermal sensor offers detailed power and temperature breakdowns of each functional unit at runtime. We also developed a thermal-aware job scheduling mechanism for reducing the performance loss due to the thermal pressure. Our methods leverage the natural discrepancies in thermal behavior among different workloads, and schedule them to keep the chip temperature within the cooling limit so as to minimize the amount of throttling. Our Linux kernel implementation of the entire framework shows noticeable performance improvements over a traditional thermal-oblivious job scheduling method while retaining its requirements for real-time and interactive jobs.

Biography of Speaker

Dr. Jun Yang received her BS from Nanjing University, PRC, her MS from University of Pittsburgh, and her Ph.D. from University of Arizona in 2002, all in computer science. Dr. Yang also holds an MS in applied mathematics from Worcester Polytechnic Institute. Dr. Yang was an assistant professor in computer science and engineering department at University of California Riverside from 2002 to 2006. She is now an assistant professor in electrical and computer engineering at Pitt. Her current research is funded mainly by National Science Foundation.