Date of Award

12-2012

Degree Type

Dissertation

Embargo Date

2-21-2013

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical Engineering and Computer Science

Advisor(s)

Qinru Qiu

Keywords

Distributed Control, Dynamic Thermal Management, Dynamic Voltage and Frequency Scaling, Low Power

Subject Categories

Electrical and Computer Engineering

Abstract

In deep submicron era, thermal hot spots and large temperature gradients significantly impact system reliability, performance, cost and leakage power. Dynamic thermal management techniques are designed to tackle the problems and control the chip temperature as well as power consumption. They refer to those techniques which enable the chip to autonomously modify the task execution and power dissipation characteristics so that lower-cost cooling solutions could be adopted while still guaranteeing safe temperature regulation. As long as the temperature is regulated, the system reliability can be improved, leakage power can be reduced and cooling system lifetime can be extended significantly.

Multimedia applications are expected to form the largest portion of workload in general purpose PC and portable devices. The ever-increasing computation intensity of multimedia applications elevates the processor temperature and consequently impairs the reliability and performance of the system. In this thesis, we propose to perform dynamic thermal management using reinforcement learning algorithm for multimedia applications. The presented learning model does not need any prior knowledge of the workload information or the system thermal and power characteristics. It learns the temperature change and workload switching patterns by observing the temperature sensor and event counters on the processor, and finds the management policy that provides good performance-thermal tradeoff during the runtime.

As the system complexity increases, it is more and more difficult to perform thermal management in a centralized manner because of state explosion and the overhead of monitoring the entire chip. In this thesis, we present a framework for distributed thermal management in many-core systems where balanced thermal profile can be achieved by proactive task migration among neighboring cores. The framework has a low cost agent residing in each core that observes the local workload and temperature and communicates with its nearest neighbor for task migration and exchange. By choosing only those migration requests that will result in balanced workload without generating thermal emergency, the presented framework maintains workload balance across the system and avoids unnecessary migration. Experimental results show that, our distributed management policy achieves almost the same performance as a global management policy when the tasks are initially randomly distributed. Compared with existing proactive task migration technique, our approach generates less hotspot, less migration overhead with negligible performance overhead.

Temperature affects the leakage power and cooling power. In this thesis, we address the impact of task allocation on a processor's leakage power and cooling fan power. Although the leakage power is determined by the average die temperature and the fan power is determined by the peak temperature, our analysis shows that the overall power can be minimized if a task allocation with minimum peak temperature is adopted together with an intelligent fan speed adjustment technique that finds the optimal tradeoff between fan power and leakage power. We further present a multi-agent distributed task migration technique that searches for the best task allocation during runtime. By choosing only those migration requests that will result chip maximum temperature reduction, the presented framework achieves large fan power savings as well as overall power reduction.

Access

Open Access

Share

COinS