Dynamic Processor Overclocking for Improving Performance of Power-Constrained Systems

Power and thermal constraints are proving to be limiting factors for computer system performance. Conventional approaches estimate a design power and cooling limit based on consumption for a specific, worst-case workload while operating at a nominal frequency. The system is then constrained to operate at or below this frequency even for workloads that have lower power and cooling requirements. As a consequence real workloads pay a performance cost for operating at a design point defined by the worst-case workloads. In this work, we propose Dynamic Processor Overclocking as a way of reducing the performance impact of the chosen design point for workloads that have sufficient slack in their power and cooling requirements. Dynamic Processor Overclocking utilizes the Dynamic Voltage and Frequency Scaling (DVS) capabilities of processors to obtain increased performance in power-constrained systems. While DVS has been widely studied in recent years, solutions exploiting it have focused on saving energy. In contrast, our proposal incorporates DVS for performance enhancement.

The paper focuses on the evaluation of the effectiveness of Dynamic Processor Overclocking using the SPEC CPU2000 benchmarks as a representative set of programs. We identify the workload characteristics that impact the benefit obtained from overclocking. We also investigate the relationship between the severity of the power constraints and application characteristics that determine the overclocking benefits. Our work presents a new approach to improving performance in power-constrained systems, and the insight and methodology required for analyzing its effectiveness for interesting workloads.

By: Juan Rubio; Karthick Rajamani; Freeman Rawson; Heather Hanson; Soraya Ghiasi; Tom Keller

Published in: RC23666 in 2005


This Research Report is available. This report has been submitted for publication outside of IBM and will probably be copyrighted if accepted for publication. It has been issued as a Research Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g., payment of royalties). I have read and understand this notice and am a member of the scientific community outside or inside of IBM seeking a single copy only.


Questions about this service can be mailed to reports@us.ibm.com .