Workload Change Point Detection for Runtime Thermal Management of Embedded Systems

Abstract

Abstract-Applications executed on multicore embedded systems interact with system software (such as the OS) and hardware, leading to widely varying thermal profiles which accelerate some aging mechanisms, reducing the lifetime reliability. Effectively managing the temperature therefore requires (1) autonomous detection of changes in application workload and (2) appropriate selection of control levers to manage thermal profiles of these workloads. In this paper we propose a technique for workload change detection using density ratio-based statistical divergence between overlapping sliding windows of CPU performance statistics. This is integrated in a run-time approach for thermal management, which uses reinforcement learning to select workload-specific thermal control levers by sampling on-board thermal sensors. Identified control levers override the OS's native thread allocation decision and scale hardware voltage-frequency to improve average temperature, peak temperature and thermal cycling. The proposed approach is validated through its implementation as a hierarchical run-time manager for Linux, with heuristic-based thread affinity selected from the upper hierarchy to reduce thermal cycling and learningbased voltage-frequency selected from the lower hierarchy to reduce average and peak temperatures. Experiments conducted with mobile, embedded and high performance applications on ARM-based embedded systems demonstrate that the proposed approach increases workload change detection accuracy by an average 3.4x, reducing the average temperature by 4-25• C, peak temperature by 6-24• C and thermal cycling by 7-35% over stateof-the-art approaches.

Real-time task scheduling and network device security for complex embedded systems based on deep learning networks

“…Recent research efforts have focused on how to improve this architecture. Methods include: enabling more performance counters to build complex system features [19] [20] [27], using powerful models for prediction [19] [18] [20] [21] [24], designing better control rules [19] [18] [20] [21], or learning control policy based on reinforcement learning [24] [25] [26] [27] [30].…”

Implementing Spiking Neural Networks on Neuromorphic Architectures: A Review

“…Using this framework, authors propose to distribute neurons and synapses to different crossbars of a hardware such that the average temperature of different crossbars is reduced, which in turn improves reliability [157]. The thermal formulation incorporates both the temporal component, resulting from self-heating of a PCM cell over time due to propagating spikes of a machine learning workload and the spatial component, resulting from heat transfer from nearby cells within a crossbar [158,159,160,161,162,163,164]. The thermal formulation is integrated inside a Hill-Climbing heuristic, which is used to map neurons and synapses to different crossbars of a hardware.…”

Optimal Planning of Dynamic Thermal Management for NANS (N-App N-Screen) Services

“…Each study in [6,8,13,15,25,26] considered different heat sources, but all studies use thermal prediction methods in their proactive DTM. Das et al [27,28] reported that their proactive run-time manager using reinforcement learning reduces thermal overheads, such as average temperature, peak temperature, and thermal cycling. The authors of [9,11] proposed DTM techniques using the power budget given by thermal prediction.…”

“…The constraint in Equation (28) says that each application's QoS should be guaranteed as higher than the minimum user required QoS u min a i at all times. The constraint in Equation (29) constrains that each application a i should be mapped to only one core out of {c 2 , c 3 , • • • , c C }.…”