Overhead Control in DP-Fair Work Conserving Real-Time Multiprocessor Scheduling

摘要

In real-time multiprocessor scheduling, the optimal global scheduling algorithms are criticized for excessive overhead due to the frequent scheduling points, migrations and preemptions. DP-Fair model is an optimal scheduling which combines the notion of fluid scheduling (ideal fairness) with deadline partitioning. It has lower number of scheduling points as compare to PFair which is the first optimal scheduling algorithm proposed for real-time multiprocessor systems. DP-Fair model exists both for non-work conserving as well as work conserving cases. In [14,15], we used some simple heuristics which lower the overhead by reducing the number of migrations and preemptions in the non-work conserving context. In this article, we show that the very same heuristics can be envisaged in case of work conserving scheduling, and we evaluate their efficiencies for lowering the overhead.