Performance evaluation of real-time multiprocessor scheduling algorithms

摘要

In this paper, we present a performance evaluation of two real-time multiprocessor scheduling algorithms: Largest Remaining Execution-Time and Local time domain (LRE-TL) and Unfair Semi-Greedy (USG). The aim of this evolution is to reflect the unseen time overhead incurred by optimal real-time algorithm, represented by LRE-TL, which might hinder the claimed optimality of such algorithms when they are practically implemented. We hyave used the CPU profiler of Oracle JavaTM VisualVM to monitor the execution of LRE-TL as well as USG algorithms. The CPU profiler of JavaTM VisualVM measures the number of invocations of scheduling event handlers (procedures) in each algorithm as well as the total time spent in all invocations of this handler. The results of the CPU profiler showed that USG outperforms LRE-TL in-terms of the number of invocations of the scheduling handlers used by each algorithm as well as the time spent by the invocations. These results suggests that optimal algorithms may turn to be non-optimal when practically implemented, unlike USG which reveals far less scheduling overhead and hence could be practically implemented in real-world applications.