Workload-aware adaptive scheduling algorithms for heterogeneous multi-core systems

Abstract

Microprocessors have undergone a revolutionary transformation since their inception, with transistor count exploding from 2,300 in the Intel 4004 to a 90 billion in the Apple M3 Ultra. This surge in processing power coexisted with challenges like performance limitations and power constraints. Multi-core processors emerged as a solution, but the "dark silicon era" highlighted the inability to fully utilize all available cores due to thermal limitations. Heterogeneous Multi-Core Processors (AMPs) addressed this by incorporating cores with diverse capabilities, maximizing power efficiency. AMPs are now all-over in the latest smartphones, tablets, and laptops. Still, the software and scheduler support remain questionable on the AMPs. Existing schedulers, designed for homogeneous systems, struggle to manage the varied capabilities of AMP cores. This research tackles this challenge by proposing a Workload-Aware Adaptive Scheduling (WAAS) algorithm. This novel approach leverages machine learning to create a workload-aware scheduler specifically designed for heterogeneous multi-core systems, aiming to unlock the full potential of AMPs and bridge the gap between processor capabilities and scheduling efficiency.