Reducing Overheating-Induced Failures Via Performance-Aware CPU Power Management

Cluster end-users and administrators have become more cognizant of the fact that large-scale commodity clusters fail quite frequently, and the main source of these failures is hardware (e.g., processors) with the primary cause being heat. This situation is expected to worsen with even larger-scale clusters powered by faster (and/or multicore) processors. In this paper, we propose a power-management algorithm that addresses heat-related reliability for processors by controlling their clock speeds in a performance-aware manner. This approach is complementary to existing approaches such as exotic cooling and faulttolerant technologies in that it proactively deals with power and cooling issues before they become a problem. Our preliminary experimental work demonstrates that our approach can easily be applied commodity processors and can reduce heat generation by 30% on average with minimal effect on performance when running the SPEC benchmarks.