DRL-based Task Scheduling and Shared Resource Allocation for Multi-Core Real-Time Systems

The worst-case execution time (WCET) of tasks running on a multi-core system is largely affected by the competition over shared resources, such as memory bandwidth and last-level cache (LLC). Therefore, the system can benefit from task scheduling and resource allocation methods that make effective use of shared resources, in terms of improving system schedulability. However, most existing scheduling and allocation methods only consider one single type of resource, e.g., memory bandwidth and LLC, and rely on static analysis that requires lower-level task details, e.g., instruction and data usage. This undermines both the applicability and effectiveness of these methods on ever-complex multi-core real-time systems. To tackle this issue, this paper proposes a Holistic Offline task Scheduling and Shared resource allocation (HOSS) framework based on deep reinforcement learning (DRL). Given a set of tasks and the underlying hardware platform, HOSS provides the collective solution of task scheduling as well as memory bandwidth and LLC allocation. The experimental results indicate that the proposed method can effectively improve system schedulability with low scheduling overhead.