2020/08/21 by Sumit K. Mandal, Mandal, Sumit K., Ümit Y. Ogras +9
Computer Science · Engineering · #Advanced Wireless Network Optimization #Artificial Intelligence (cs.AI) #Caching and Content Delivery #Distributed #FOS: Computer and information sciences #FOS: Electrical engineering #Machine Learning (cs.LG) #Parallel #Parallel Computing and Optimization Techniques #Systems and Control (eess.SY) #and Cluster Computing (cs.DC) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2008.09728
openalex publication_date 2020/08/21 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Dynamic resource management has become one of the major areas of research in\nmodern computer and communication system design due to lower power consumption\nand higher performance demands. The number of integrated cores, level of\nheterogeneity and amount of control knobs increase steadily. As a result, the\nsystem complexity is increasing faster than our ability to optimize and\ndynamically manage the resources. Moreover, offline approaches are sub-optimal\ndue to workload variations and large volume of new applications unknown at\ndesign time. This paper first reviews recent online learning techniques for\npredicting system performance, power, and temperature. Then, we describe the\nuse of predictive models for online control using two modern approaches:\nimitation learning (IL) and an explicit nonlinear model predictive control\n(NMPC). Evaluations on a commercial mobile platform with 16 benchmarks show\nthat the IL approach successfully adapts the control policy to unknown\napplications. The explicit NMPC provides 25% energy savings compared to a\nstate-of-the-art algorithm for multi-variable power management of modern GPU\nsub-systems.\n