FRAMEWORK FOR IMPLEMENTING A MACHINE LEARNING WORKFLOW IN DYNAMIC VOLTAGE AND FREQUENCY SCALING (DVFS) FOR IMPROVED POWER AND THERMAL MANAGEMENT
Keywords:
Dynamic Voltage Frequency Scaling (DFVS), Deep Reinforcement Learning, Machine Learning, Power And Thermal Management, MicroprocessorAbstract
The ongoing advancement of technology calls for continuous enhancement of energy efficiency in microprocessors. One of the fundamental techniques employed for such improvements is Dynamic Voltage and Frequency Scaling (DVFS), which strategically adjusts power levels based on varying workloads. However, finding the optimal balance for these adjustments presents a complex challenge that needs to be addressed. This paper proposes a novel approach that utilizes Machine Learning (ML) models to enhance the DVFS design and simulation framework. The proposed framework incorporates three critical modules- microarchitectural simulation, ML model-based prediction, and Deep Reinforcement Learning (DRL) based DVFS control—offering an intelligent management mechanism that efficiently trades off among different optimization targets. The paper discusses various ML models and their effectiveness in predicting workload patterns, enabling precise estimation of power and thermal dissipation.
References
Zhuo, C., Gao, D., Cao, Y., Shen, T., Zhang, L., Zhou, J., & Yin, X. (2021). A DVFS Design and Simulation Framework Using Machine Learning Models. IEEE Design & Test.
Ranjbar, B., Klemme, F., Genssler, P. R., Amrouch, H., Jung, J., Dave, S., ... & Kumar, A. (2023, April). Learning-Oriented Reliability Improvement of Computing Systems From Transistor to Application Level. In 2023 Design, Automation & Test in Europe Conference & Exhibition (DATE) (pp. 1-10). IEEE. [3] Zhuo, C., Gao, D., Cao, Y., Shen, T., Zhang, L., Zhou, J., & Yin, X. (2021). A DVFS Design and Simulation Framework Using Machine Learning Models. IEEE Design & Test.
Gupta, M., Bhargava, L., & Indu, S. (2021). Mapping techniques in multicore processors: current and future trends. The Journal of Supercomputing, 77, 9308-9363.
Fettes, Q., Clark, M., Bunescu, R., Karanth, A., & Louri, A. (2018). Dynamic voltage and frequency scaling in NoCs with supervised and reinforcement learning techniques. IEEE Transactions on Computers, 68(3), 375-389.
Rogers, S., Slycord, J., Baharani, M., & Tabkhi, H. (2020, October). gem5-salam: A system architecture for llvm-based accelerator modeling. In 2020 53rd Annual IEEE/ACM International Symposium on Microarchitecture (MICRO) (pp. 471-482). IEEE.
Zhang, K. (2020). Task Offloading and Resource Allocation using Deep Reinforcement Learning (Doctoral dissertation, Université d'Ottawa/University of Ottawa).
Wu, Q., Reddi, V. J., Wu, Y., Lee, J., Connors, D., Brooks, D., ... & Clark, D. W. (2005, November). A dynamic compilation framework for controlling microprocessor energy and performance. In 38th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO'05) (pp. 12-pp). IEEE.
Downloads
Published
Issue
Section
License
Copyright (c) 2023 Annewsha Datta, Rudrendu Kumar Paul, Aryyama Kumar Jana (Author)
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
