Exploiting task temperature profiling in temperature-aware task scheduling for computational clusters

Authors:
Daniel C. Vanderster;Amirali Baniasadi;Nikitas J. Dimopoulos
Affiliations:
Department of Electrical and Computer Engineering, University of Victoria, Victoria, Canada;Department of Electrical and Computer Engineering, University of Victoria, Victoria, Canada;Department of Electrical and Computer Engineering, University of Victoria, Victoria, Canada
Venue:
ACSAC'07 Proceedings of the 12th Asia-Pacific conference on Advances in Computer Systems Architecture
Year:
2007

Citing 6
Cited 1

Design Challenges of Technology Scaling

IEEE Micro
Simgrid: A Toolkit for the Simulation of Application Scheduling

CCGRID '01 Proceedings of the 1st International Symposium on Cluster Computing and the Grid
The Impact of Technology Scaling on Lifetime Reliability

DSN '04 Proceedings of the 2004 International Conference on Dependable Systems and Networks
Power and Energy Management for Server Systems

Computer
Resource Management and Knapsack Formulations on the Grid

GRID '04 Proceedings of the 5th IEEE/ACM International Workshop on Grid Computing
Making scheduling "cool": temperature-aware workload placement in data centers

ATEC '05 Proceedings of the annual conference on USENIX Annual Technical Conference

Review of performance metrics for green data centers: a taxonomy study

The Journal of Supercomputing

Quantified Score

Hi-index	0.00

Visualization

Abstract

Many years of CMOS technology scaling have resulted in increased power densities and higher core temperatures. Power and temperature concerns are now considered to be a primary challenge for continued scaling and long-term processor reliability. While solutions for low-power and low-temperature circuits and microarchitectures have been studied for many years, temperature-awareness at the computational cluster level is a relatively new problem. To address this problem, we introduce a temperature-aware task scheduler based on task temperature profiling. We study the task characteristics and temperature profiles for a subset of SPEC'2K benchmarks. We exploit these profiles and suggest several scheduling algorithms aimed at achieving lower cluster temperature. Our findings show a clear trade-off between the overall queue servicing time and the cluster peak temperature. Whether the temperature reductions achieved are worth the extra delay is left to the designer/user to decide based on the case by case performance restrictions and temperature limitations.