Machine learning for run-time energy optimisation in many-core systems

Abstract

Abstract-In recent years, the focus of computing has moved away from performance-centric serial computation to energyefficient parallel computation. This necessitates run-time optimisation techniques to address the dynamic resource requirements of different applications on many-core architectures. In this paper, we report on intelligent run-time algorithms which have been experimentally validated for managing energy and application performance in many-core embedded system. The algorithms are underpinned by a crosslayer system approach where the hardware, system software and application layers work together to optimise the energyperformance trade-off. Algorithm development is motivated by the biological process of how a human brain (acting as an agent) interacts with the external environment (system) changing their respective states over time. This leads to a pay-off for the action taken, and the agent eventually learns to take the optimal/best decisions in future. In particular, our online approach uses a model-free reinforcement learning algorithm that suitably selects the appropriate voltage-frequency scaling based on workload prediction to meet the applications' performance requirements and achieve energy savings of up to 16% in comparison to stateof-the-art-techniques, when tested on four ARM A15 cores of an ODROID-XU3 platform.

Learning-Oriented QoS- and Drop-Aware Task Scheduling for Mixed-Criticality Systems

“…Learning Action Determination: There are various methods for determining the optimal action a t according to the state s t . In this paper, the well-known -greedy policy has been exploited, in which the dynamic policy is used for adjusting [36]. In -greedy policy, random action is selected from the actions set with the probability of , i.e., the best action is selected with the largest Q-value with the probability of 1 − .…”

A Survey of Prediction and Classification Techniques in Multicore Processor Systems

“…An online DVFS control strategy based on core-level modular reinforcement learning to adaptively select appropriate operating frequencies for each individual core was proposed in [60]. An Q-learning based algorithm was proposed in [61] to identify V/F pairs for predicted workloads and given application performance requirements. The study in [62] investigated imitation learning and reported higher quality policies in the context of dynamic VFI control in many core systems with different applications running concurrently.…”

Empirical model-based performance prediction for application mapping on multicore architectures

“…Then, cover a number of approaches relying on reinforcement learning and supervised learning. In [18], reinforcement learning is applied through a cross-layer system approach to predict the best energy-performance trade-off in multicore embedded systems. It relies on a biologically-inspired runtime power management framework implementing a Q-learning algorithm, which selects the voltage-frequency levels to minimize energy consumption.…”