Performance and Reliability-Driven Scheduling Approach for Efficient Execution of Parallelizable Stochastic Tasks in Heterogeneous Computing Systems

In heterogeneous computing systems, a suite of different machines is interconnected to provide a variety of computational capabilities to execute collections of real-time application tasks that have diverse requirements. In these systems, machine failures are inevitable and can have an adverse effect on applications executing on the system. To reduce this effect, an efficient scheduling approach that is capable to assign tasks of an application to the suitable processors in these systems, to yield high performance and reliability of the system, must be devised. In this paper, we investigate a non-preemptive heuristic dynamic scheduling scheme for a set of independent stochastic tasks with parallelizable contents executing on a heterogeneous computing environment. Each stochastic task is characterized by its deadline and its time cost distribution, which is given in terms of mean and variance. A scheduled task may run as one unit on a single processor, or parallelized and its split tasks can be executed concurrently on multiple processors. We use this parallelism in tasks to meet their deadlines and also obtain better processor utilization compared to non-parallelized tasks. Non-preemptive parallelizable task scheduling combines the advantages of higher schedulability and lower scheduling overhead offered by the preemptive and non-preemptive task scheduling models, respectively. The scheduling algorithm proposed in this paper takes reliability measures into account, thereby enhancing the reliability of heterogeneous systems without any Ehab Y. Abdel Maksoud 138 additional hardware cost. To make scheduling results more realistic and precise, we exploits parallelism in tasks, if it is necessary, and distributes them among the available processors to maximize both of the reliability of the application and the performance measures of the system, such as the schedulability and the processor utilization, while maintaining lower scheduling overhead at the same time, compared to the previous task scheduling approaches.