HiDVFS Multi-Agent Scheduler Cuts Energy 55% with 4.6x Speedup
New hierarchical AI scheduler slashes energy and boosts performance in embedded multicore systems
Deep Dive
HiDVFS, a hierarchical multi-agent DVFS scheduler proposed for real-time OpenMP DAG workloads on embedded multicore systems, uses profiler, thermal, and priority agents to respect deadlines and thermal limits. On Jetson TX2, it achieves a 2.83x speedup and 32.9% energy reduction over a fairness-corrected GearDVFS port, and across all 12 BOTS benchmarks a 4.62x average speedup with 55.7% energy reduction.
Key Points
- 4.62x average speedup and 55.7% energy reduction across 12 BOTS benchmarks
- Hierarchical multi-agent architecture uses profiler, thermal, and priority agents for per-core DVFS control
- Cross-platform validation on Jetson TX2, Orin NX, and RubikPi shows 15-18% energy savings over max-frequency pinning
Why It Matters
Enables power-efficient real-time AI inference and control on embedded multicore systems, crucial for edge and IoT deployments.