Research On Dependent Task Scheduling Strategy Based On Heterogeneous Multi-core Processors

With the increasingly concern on high performance computing field in native and abroad,processor GHz and chip integration were greatly enhanced. As the limitation of the chip area and gradually approaching the upper limit of integration, the processor performance reaches the bottlenecks. Based on the status and future trends of processor, multi-core processors has appeared on the history stage and rapidly rise to the international mainstream processing architecture. Hardware platform and suitable software is a necessary condition for the full release of system performance. With the more and more deeply research to the related technology of multi-core processor, the expert of high performance computing field has found out that task scheduling strategy is one of the important factors to the full release of system performance.Based on the deeply research and detailed analysis to the characteristics of heterogeneous platforms and task scheduling algorithm, the paper proposed an efficient algorithm that dependent task scheduling strategy based on heterogeneous multi-core processors which takes an improved list priority task scheduling algorithm as main part. The algorithm proposed in this paper contains a priority sorting method that can suit the heterogeneous platform features and dependent task character well, which takes priority list scheduling algorithm as the basic guiding theory and overcoming the shortage that takes the average computing time of task as the priority parameter. On this basis, the algorithm has improved the pattern that use the greedy theory mapping task to the processor which may bring the local optimal solutions, and proposed a new mapping method based on task backward critical path to achieve the global optimal solution. Meanwhile, in the task assignment stage, the algorithm uses multi-duplication technique to earlier the execution time,and finally uses interval inserted technique assigned tasks to processing cores for improving the utilization of processor.At last, to prove that the algorithm is an efficient algorithm to solve the problem in the researching field, by using reasonable performance evaluation criteria and generating random DAG task graph to obtain test cases, the paper proposed performance verification program and did some simulation tests on Simics simulation platform. The experimental data was analyzed by a comparison way. Performance test results show that: the algorithm proposed in this paper can efficient solve the current problems that the critical path cannot scheduling first and scheduling result falling into local optimal solution etc. The algorithm can improve system performance.