HiDVFS: A Hierarchical Multi-Agent DVFS Scheduler for OpenMP DAG Workloads

TL;DR

The paper tackles leakage power and thermal challenges in embedded multicore systems by focusing on energy- and temperature-aware DVFS scheduling for OpenMP DAG workloads. It introduces HiDVFS, a hierarchical multi-agent reinforcement learning scheduler with profiler, thermal, and priority agents that optimize per-core frequencies, core allocations, and DAG priorities under a makespan-first objective, aided by model-based reward estimation for few-shot learning. Key contributions include decomposing the action space to improve sample efficiency, formalizing reward functions, and validating on NVIDIA Jetson TX2 with 9 BOTS benchmarks, achieving substantial improvements in makespan and energy (e.g., average speedups around 3.95x and energy reductions near 47%). The work demonstrates practical viability of fine-grained DVFS control in irregular OpenMP DAG workloads, enabling energy savings and thermal safety in embedded real-time systems, with robust cross-seed performance on real hardware.

Abstract

With advancements in multicore embedded systems, leakage power, exponentially tied to chip temperature, has surpassed dynamic power consumption. Energy-aware solutions use dynamic voltage and frequency scaling (DVFS) to mitigate overheating in performance-intensive scenarios, while software approaches allocate high-utilization tasks across core configurations in parallel systems to reduce power. However, existing heuristics lack per-core frequency monitoring, failing to address overheating from uneven core activity, and task assignments without detailed profiling overlook irregular execution patterns. We target OpenMP DAG workloads. Because makespan, energy, and thermal goals often conflict within a single benchmark, this work prioritizes performance (makespan) while reporting energy and thermal as secondary outcomes. To overcome these issues, we propose HiDVFS (a hierarchical multi-agent, performance-aware DVFS scheduler) for parallel systems that optimizes task allocation based on profiling data, core temperatures, and makespan-first objectives. It employs three agents: one selects cores and frequencies using profiler data, another manages core combinations via temperature sensors, and a third sets task priorities during resource contention. A makespan-focused reward with energy and temperature regularizers estimates future states and enhances sample efficiency. Experiments on the NVIDIA Jetson TX2 using the BOTS suite (9 benchmarks) compare HiDVFS against state-of-the-art approaches. With multi-seed validation (seeds 42, 123, 456), HiDVFS achieves the best finetuned performance with 4.16 plus/minus 0.58s average makespan (L10), representing a 3.44x speedup over GearDVFS (14.32 plus/minus 2.61s) and 50.4% energy reduction (63.7 kJ vs 128.4 kJ). Across all BOTS benchmarks, HiDVFS achieves an average 3.95x speedup and 47.1% energy reduction.

Figures (18)

• Figure 1: OpenMP DAG snippet showing tied/untied tasks and dependency-induced variability in execution time.
• Figure 2: Monitoring one user DAG ($\tau_1$): five jobs ($J_{1,1,1}$--$J_{3,1,4}$) mapped to cores ($c_1$--$c_3$) with target frequencies ($f_0$--$f_2$). Core $c_0$ runs system DAG ($\tau_0$).
• Figure 3: N-Queens on Xeon (12 cores) and Core i7 (4 cores). More cores increase branch misses for tied tasks, raising makespan and energy. Shaded regions show variation over 10 runs; serial reduces misses.
• Figure 4: Scenarios on a four-core system. Blue: performance; green: powersave. $PP, PS$ denote unbounded parallel execution; $SP, SS$ denote bounded serial execution. The green dashed line marks DAG makespan. HiDVFS agents (7–9) align core choice, frequency, and temperature.
• Figure 5: Importance of different features on total energy consumption and makespan in parallel and sequential execution of parallel applications.