Unleashing the Potential of Asymmetric Multicore Processors Through Operating System Support05/28/2009
Speaker(s) : Sasha Fedorova (SFU)
Asymmetric multicore processors (AMP) promise higher performance per watt than their symmetric counterparts, and it is likely that future processors will integrate a few fast out-of-order cores, coupled with a large number of simpler, slower cores, all exposing the same ISA. Abundant simple cores enable scalability for highly parallel applications, while complex and powerful cores deliver good performance for sequential applications as well as sequential phases of parallel applications. To fully tap into the potential of AMP systems, the operating system must map threads to asymmetric cores in consideration of the amount of parallelism in the application. Our work is the first to design, implement and evaluate such operating system support. Our new Parallelism-Aware (PA) scheduling algorithm maps applications to cores on an AMP according to the degree of dynamic parallelism in the application, but without compromising the fairness and load balance of the system. Using our implementation we are able to evaluate the feasibility of delivering the benefits of AMP to applications through operating system support, without requiring changes to applications or their recompilation. We show that the PA algorithm accomplishes this goal, delivering as much as 27% better performance than the asymmetry-agnostic scheduler for applications with a non-negligible sequential component. We also conclude that to avoid the overhead of cross-chip thread migrations, future AMP systems should be built such that fast and slow cores are placed within close physical proximity on the same chip. Bio: Alexandra (Sasha) Fedorova is an Assistant Professor of Computer Science at Simon Fraser University (SFU) in Vancouver, Canada. She has earned her Ph.D. at Harvard University in 2006, where she has completed a thesis on Operating System Scheduling for Chip Multithreaded Processors. Concurrently with her doctorate studies, Fedorova worked at Sun Microsystems Research Labs, where she investigated transactional memory and operating systems. She is the lead inventor on two US patents. At SFU, Fedorova has co-founded Systems, Networking and Architecture (SYNAR) research lab. Presently, her research interests span operating systems and virtualization platforms for multi-core. Recently she and her students have been working on tools and techniques for parallelization of video games. Dr. Fedorova's research is recognized by researchers and practitioners alike. Her work is supported by Sun Microsystems, Intel and Electronic Arts. She also holds a prestigious Strategic Grant from the Canadian government.
Lamia.Benmouffok (at) nulllip6.fr