CQGA-HEFT: Q-learning-based DAG Scheduling Algorithm Using Genetic Algorithm in Clustered Many-core Platform

Abstract

Embedded systems, e.g., self-driving systems, are becoming larger and more complex, and the performance requirements of such platforms are increasing. Clustered many-core processors satisfy these requirements because they provide an isolated computing environment, low power consumption, and high performance. However, efficiently scheduling many tasks to clustered many-core processors is a difficult problem. Therefore, this paper proposes a scheduling algorithm for clustered many-core processors based on Q-learning (a reinforcement learning method) that is effective for scheduling many tasks. The proposed algorithm distinguishes the difference in communication time between bus communication and network-on-chip communication for a clustered many-core processor. Therefore, in the proposed algorithm’s learning process, these two types of communication times are considered based on a genetic algorithm (GA). Comparative experiments demonstrate that the proposed algorithm outperforms existing methods in terms of makespan.