2013 Poster Sessions : Dynamic Management of TurboMode in Modern Multi-core Chips

Student Name : David Lo
Advisor : Christos Kozyrakis
Research Areas: Computer Systems
Dyanmic overclocking of CPUs, or TurboMode, is a recently introduced feature on many x86 multi-core processors that can improve the performance of applications. However, naïve use of TurboMode may lead to energy inefficiency. Thus far, there is no strategy for managing TurboMode effectively across workloads and efficiency metrics. This paper analyzes the impact of TurboMode on several efficiency metrics for server workloads and proposes autotuner, an automatic management scheme for TurboMode. We demonstrate that autotuner can improve QPS/$, ED, and ED^2 by 8%, 47%, and 68% over naïve use of TurboMode.

David Lo is a 2nd year EE PhD student working for Professor Christos Kozyrakis. His primary research area is computer architecture. He has previously researched scalable programming models for multicore systems. His current research focuses on increasing energy efficiency of datacenters through better resource utilization and coordination between hardware and software.