一种基于DVFS特性曲线的异构计算单元低功耗协同调度方法

A Low-Power Cooperative Scheduling Method for Heterogeneous Computing Units Based on DVFS Characteristic Curves

摘要: 针对工业边缘计算环境中任务动态到达、实时性要求高、资源受限且能耗敏感的问题，本文提出一种基于DVFS特性曲线的异构计算单元低功耗协同调度方法(简称：CS-HCU-DVFS)。该方法融合任务调度与硬件功耗管理，实现调度决策与能效调控的协同优化。框架由任务特征分析、能效感知调度引擎和DVFS协同调控模块组成：任务特征分析模块提取任务类型、数据量及截止时间；调度引擎结合异构单元的功耗–性能特性表(PPCT)，以能效比最大化为目标，查表确定最优计算单元与运行频率；DVFS模块据此动态配置电压与时钟频率，并通过抑制频繁切换提升系统稳定性。执行完成后，计算单元反馈状态信息至调度引擎，形成闭环调控机制。实验结果表明，该方法在满足任务实时性约束的前提下，有效降低系统能耗，提升整体能效。该方法调度开销低，适用于资源受限的工业边缘计算场景。

Abstract: To address the challenges of dynamically arriving tasks, high real-time requirements, resource constraints, and energy sensitivity in industrial edge computing environments, this paper proposes a low-power cooperative scheduling method for heterogeneous computing units based on DVFS characteristic curves, named CS-HCU-DVFS. The method integrates task scheduling with hardware power management to achieve optimization of scheduling decisions and energy-efficiency control. The framework consists of three components: task feature analysis, energy-aware scheduling engine, and DVFS cooperative control module. The task feature analysis module extracts task type, data volume, and deadline constraints. The scheduling engine leverages a power-performance characteristic table (PPCT) of heterogeneous units to determine the optimal computing unit and operating frequency by table lookup, aiming to maximize energy efficiency. The DVFS module dynamically configures voltage and clock frequency accordingly, while suppressing frequent transitions to enhance system stability. Upon task completion, computing units feedback status information to the scheduling engine, forming a closed-loop control mechanism. Experimental results show that the proposed method effectively reduces system energy consumption and improves overall energy efficiency while meeting real-time constraints. With low scheduling overhead, CS-HCU-DVFS is suitable for resource-constrained industrial edge computing scenarios.