Tuesday Nov. 30, 2010
Location: Hamerschlag Hall D-210
Carnegie Mellon University
In a modern chip-multiprocessor system, memory is a shared resource among multiple concurrently executing threads. The memory scheduling algorithm should resolve memory contention by arbitrating accesses in such a way that competing threads progress at a relatively fast and even pace, resulting in high system throughput and fairness. However, previously proposed memory scheduling algorithms are predominantly optimized for only one of these objectives. The Thread Cluster Memory Scheduler (TCM) addresses system throughput and fairness separately with the goal of achieving the best of both. It dynamically groups threads with similar memory access behavior into either the latency-sensitive or the bandwidth-sensitive cluster. Latency-sensitive threads are “light” and can significantly contribute to system throughput when prioritized. Bandwidth-sensitive threads are “heavy” and can potentially starve each other out, resulting in unfairness. TCM employs disparate scheduling policies for each cluster to address their needs separately and simultaneously.Evaluations show that TCM outperforms previously proposed schedulers with respect to both system throughput and fairness.
Yoongu Kim is a third year Ph.D. student in the Electrical and Computer Engineering department. He received his B.S. in Electrical Engineering from Seoul National University in 2005. He researches better ways to resolve contention for shared system resources in multicore architectures. He works closely with Professors Onur Mutlu and Mor Harchol-Balter.