SITSA-RT: An Information Theory Inspired Real-Time Multiprocessor Scheduler

Abstract

In this paper, we describe how Shannon's information theory is used to develop the Simplified Information-Theoretic Scheduling algorithm for Real-time Systems (SITSA-RT), and we explain the mechanism used by this algorithm to reduce the number of job migrations in real-time systems implemented in a multiprocessor platform. We present a performance comparison of the proposed algorithm with different multiprocessor scheduling algorithms for synthetic and real-case task sets. The results of the performance comparison for the synthetic task sets case show that outperforms all the studied EDF-based (up to 41.65%) and P-Fair based algorithms (up to 93.22%) in terms of the reduction of the number of job migrations while offering a similar performance in terms of the number of preemptions, the number of tasks migrations, and deadline miss ratio. These results show that as the utilization per task set and the number of processors increase, SITSA-RT is able to improve its performance in terms of the number of migrations. The results from the real-case task set based on NASA's X-38 avionics architecture show that for the scheduler execution time, MLLF improves the performance of SITSA-RT by 5.96% and SITSA-RT improves the performance of LLF by 19%, and from the memory requirements we found that MLLF usage is 13.48% lower than SITSA-RT, and SITSA-RT usage is 52.97% lower than LLF.