Research On Dependent Tasks Scheduling Of Heterogeneous CMP Based On Improved Particle Swarm

With the generation and application of multi-core processors, its operational efficiencyand parallel performance obtain much attention of experts and scholars as well as the majorhardware manufacturers at home and abroad. The system performance is not merely improvesby advancing in hardware platform technology, but also needs their hardware platform withoptimization software in order to give full play to the excellent performance of the technologyadvances. Therefore, task scheduling in multi-core processors has become one of the hotissues of high-performance parallel processor. In view of the excellent performance andprospects for the development of multi-core processor platforms, to further enhance its actualapplication performance, a dependent task scheduling algorithm suitable to heterogeneousplatform with excellent performance is designed to shorten the time of the implementation ofthe completion of all tasks, and to improve the operating efficiency of the multi-coreprocessor systems.In this paper, by the analysis of multi-core processors and its existing task schedulingalgorithm, drawing on the idea of particle swarm optimization algorithm, by improved andoptimized it, and then propose a dependent task scheduling algorithms on heterogeneousmulticore processor platforms. The algorithm uses the idea of random parallel search based onparticle swarm optimization to research for the optimal solution of the task schedulingproblem in the particle space, and Chaos theory was introduced into particle swarmoptimization population initialization and local search to improve the local search ability ofthe particle swarm algorithm. According to the characteristics of heterogeneous multicoresystem dependent task scheduling, a new particle encoding and decoding programs, updateguidelines are designed based on the characteristics of the task scheduling problem, and thenot feasible scheduling solution is converted to the better feasible solutions under the guide ofthe dual heuristic priority rules, while incorporating adaptive learning strategymulti-neighborhood search to further enhance the speed of convergence of the algorithm andthe local fine search performance. The algorithm can find the in optimal task schedulingprogram in a short period of time to reduce the time of the completion of all tasks execute andenhance the performance of the parallelism of the heterogeneous multi-core processors.In order to prove the feasibility and efficiency of the task algorithms on heterogeneoussystem, experiments using randomly generated set of task graph into three categories with different characteristics and different hardware conditions that the processor core number asthe algorithm test input simulation to achieve convergence speed and performance of thealgorithm in Matlab platform. Experimental results show that: Compared with existingalgorithms, the new algorithm can always find the optimal or suboptimal solutions fordependent tasks scheduling problems in heterogeneous multi-core processor, that the timewhen the execution of all the tasks were completed was the shortest, and it has good stabilityand scalability, and at the same time can provide a good reference for related research.