QoS-Efficient and Near-Optimal Imprecise Computation Task Deployment for Heterogeneous Multicore Systems

Heterogeneous multicore platforms are widely used in many applications due to their customizability, where energy efficiency, real-time response, and system quality-of-service (QoS) are critical. Imprecise computation (IC) tasks provide an efficient architecture to balance these conflicting requirements. Existing methods seldom consider deploying IC tasks on heterogeneous platforms under complex scenarios and providing an efficient way to find proper solutions. This paper studies the joint-design problem of mapping and scheduling dependent IC tasks on heterogeneous multicores. The target problem is formulated as a mixed-integer nonlinear program (MINLP) that maximizes system QoS under energy supply, task deadline, precedence, and non-overlap constraints. To solve the problem efficiently, we propose a decomposition-based solution, including a master problem (MP) that handles task allocation and execution cycle adjustment, a scheduling problem (SP) that focuses on dependency-aware task scheduling, and a closed-loop feedback mechanism from SP to MP to improve task schedulability. Experimental results show that the proposed method achieves 95.25% of the optimal QoS obtained by a commercial solver while using only about 19.8% of its computation time, demonstrating a favorable trade-off between solution quality and computational efficiency.