TY - JOUR
T1 - An Energy-aware Online Learning Framework for Resource Management in Heterogeneous Platforms
AU - Mandal, Sumit K.
AU - Bhat, Ganapati
AU - Doppa, Janardhan Rao
AU - Pande, Partha Pratim
AU - Ogras, Umit Y.
N1 - Funding Information:
This work was supported partially by USA Army Research Ofce grant W911NF-17-1-0485, National Science Foundation grants CNS-1526562 and OAC-1910213, and Semiconductor Research Corporation (SRC) task 2721.001.
Funding Information:
This work was supported partially by USA Army Research Office grant W911NF-17-1-0485, National Science Foundation grants CNS-1526562 and OAC-1910213, and Semiconductor Research Corporation (SRC) task 2721.001. Authors’ addresses: S. K. Mandal, G. Bhat, and U. Y. Ogras, Arizona State University, School of Electrical, Computer, and Energy Engineering, Tempe, AZ 85287, USA; emails: {skmandal, gmbhat, umit}@asu.edu; J. R. Doppa and P. P. Pande, Washington State University, School of EECS, Pullman, WA 99164, USA; emails: {jana.doppa, pande}@wsu.edu. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from permissions@acm.org. © 2020 Association for Computing Machinery. 1084-4309/2020/05-ART28 $15.00 https://doi.org/10.1145/3386359
Publisher Copyright:
© 2020 ACM.
PY - 2020/5/7
Y1 - 2020/5/7
N2 - Mobile platforms must satisfy the contradictory requirements of fast response time and minimum energy consumption as a function of dynamically changing applications. To address this need, systems-on-chip (SoC) that are at the heart of these devices provide a variety of control knobs, such as the number of active cores and their voltage/frequency levels. Controlling these knobs optimally at runtime is challenging for two reasons. First, the large configuration space prohibits exhaustive solutions. Second, control policies designed offline are at best sub-optimal, since many potential new applications are unknown at design-time. We address these challenges by proposing an online imitation learning approach. Our key idea is to construct an offline policy and adapt it online to new applications to optimize a given metric (e.g., energy). The proposed methodology leverages the supervision enabled by power-performance models learned at runtime. We demonstrate its effectiveness on a commercial mobile platform with 16 diverse benchmarks. Our approach successfully adapts the control policy to an unknown application after executing less than 25% of its instructions.
AB - Mobile platforms must satisfy the contradictory requirements of fast response time and minimum energy consumption as a function of dynamically changing applications. To address this need, systems-on-chip (SoC) that are at the heart of these devices provide a variety of control knobs, such as the number of active cores and their voltage/frequency levels. Controlling these knobs optimally at runtime is challenging for two reasons. First, the large configuration space prohibits exhaustive solutions. Second, control policies designed offline are at best sub-optimal, since many potential new applications are unknown at design-time. We address these challenges by proposing an online imitation learning approach. Our key idea is to construct an offline policy and adapt it online to new applications to optimize a given metric (e.g., energy). The proposed methodology leverages the supervision enabled by power-performance models learned at runtime. We demonstrate its effectiveness on a commercial mobile platform with 16 diverse benchmarks. Our approach successfully adapts the control policy to an unknown application after executing less than 25% of its instructions.
KW - Dynamic power management
KW - imitation learning
KW - online learning
KW - reinforcement learning
UR - http://www.scopus.com/inward/record.url?scp=85085257238&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85085257238&partnerID=8YFLogxK
U2 - 10.1145/3386359
DO - 10.1145/3386359
M3 - Article
AN - SCOPUS:85085257238
SN - 1084-4309
VL - 25
JO - ACM Transactions on Design Automation of Electronic Systems
JF - ACM Transactions on Design Automation of Electronic Systems
IS - 3
M1 - 28
ER -