Multi-round divisible real-time scheduling algorithm on multiprocessor platforms

Abstract

Recent real-time systems and applications are becoming more complex and contain more functionality. Therefore, these systems are increasingly to be implemented upon multiprocessor platforms, as they require complex sharing of data, synchronization and parallelism. To overcome this limitation, recent researches have applied Divisible Load Theory (DLT) to real-time multiprocessor scheduling and the theory is known as Real-time Divisible Load Theory (RT-DLT). However, most current studies in this field are about distributing data in single-round algorithm and there are limited studies in multi-round strategy in real-time systems to reduce idle time. Moreover, current multi-round studies have some performance problems mainly due to inefficient use of available resources and long execution time for task scheduling. This research is carried out to address the problem of task execution on real-time multiprocessor platforms to reduce inserted idle time in order to meet task deadline. Therefore to achieve that, this research developed three significant multiround algorithms which are: MultiMINPROCS, OPTROUND and MINCOMPTIME in expanding the current single-round RT-DLT to multi-round RT-DLT. Series of experimental evaluations showed that the three developed algorithms had improved the performance of previous both single-round and multi-round algorithms. The first algorithm computed the minimum number of processors needed to complete the job by its deadline, 40% improved the previous single-round algorithm and 33% improved previous multi-round algorithm. The second algorithm determined the most efficient number of round. Finally the third algorithm computed the minimum completion time in order to meet the task’s deadline, 35% improved the previous single-round algorithm and 38% improved previous multi-round algorithm.