TY - GEN
T1 - Orchestrated Co-scheduling, Resource Partitioning, and Power Capping on CPU-GPU Heterogeneous Systems via Machine Learning
AU - Saba, Issa
AU - Arima, Eishi
AU - Liu, Dai
AU - Schulz, Martin
N1 - Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - CPU-GPU heterogeneous architectures are now commonly used in a wide variety of computing systems from mobile devices to supercomputers. Maximizing the throughput for multi-programmed workloads on such systems is indispensable as one single program typically cannot fully exploit all avaiable resources. At the same time, power consumption is a key issue and often requires optimizing power allocations to the CPU and GPU while enforcing a total power constraint, in particular when the power/thermal requirements are strict. The result is a system-wide optimization problem with several knobs. In particular we focus on (1) co-scheduling decisions, i.e., selecting programs to co-locate in a space sharing manner; (2) resource partitioning on both CPUs and GPUs; and (3) power capping on both CPUs and GPUs. We solve this problem using predictive performance modeling using machine learning in order to coordinately optimize the above knob setups. Our experiential results using a real system show that our approach achieves up to 67% of speedup compared to a time-sharing-based scheduling with a naive power capping that evenly distributes power budgets across components.
AB - CPU-GPU heterogeneous architectures are now commonly used in a wide variety of computing systems from mobile devices to supercomputers. Maximizing the throughput for multi-programmed workloads on such systems is indispensable as one single program typically cannot fully exploit all avaiable resources. At the same time, power consumption is a key issue and often requires optimizing power allocations to the CPU and GPU while enforcing a total power constraint, in particular when the power/thermal requirements are strict. The result is a system-wide optimization problem with several knobs. In particular we focus on (1) co-scheduling decisions, i.e., selecting programs to co-locate in a space sharing manner; (2) resource partitioning on both CPUs and GPUs; and (3) power capping on both CPUs and GPUs. We solve this problem using predictive performance modeling using machine learning in order to coordinately optimize the above knob setups. Our experiential results using a real system show that our approach achieves up to 67% of speedup compared to a time-sharing-based scheduling with a naive power capping that evenly distributes power budgets across components.
KW - CPU-GPU heterogeneous systems
KW - Co-scheduling
KW - Machine learning
KW - Power capping
KW - Resource partitioning
UR - http://www.scopus.com/inward/record.url?scp=85144815176&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-21867-5_4
DO - 10.1007/978-3-031-21867-5_4
M3 - Conference contribution
AN - SCOPUS:85144815176
SN - 9783031218668
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 51
EP - 67
BT - Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings
A2 - Schulz, Martin
A2 - Trinitis, Carsten
A2 - Papadopoulou, Nikela
A2 - Pionteck, Thilo
PB - Springer Science and Business Media Deutschland GmbH
T2 - 35th International Conference on Architecture of Computing Systems, ARCS 2022
Y2 - 13 September 2022 through 15 September 2022
ER -