基于改进量子粒子群的分布式并行计算框架设计

Design of Distributed Parallel Computing Framework Based on Improved Quantum Particle Swarm [HS)]

摘要:

为了实现用户任务在大规模计算机集群上进行高效地处理,并克服现有并行计算框架通用性不强的缺点,提出了一种基于改进量子群算法和Map-Reduce模型的通用并行计算框架;首先,对经典的Map-Reduce分布式并行计算框架以及并行计算流程进行了具体描述;然后,基于改进的量子粒子群算法设计了改进的Map-Reduce模型,在Map阶段通过多种群并行搜索并计算所有粒子适应度,在Shuffle和Sort 阶段实现粒子的排序和种群的重新划分,然后在Reduce阶段更新控制系数和粒子位置,当最优解不变时,通过混沌扰动对其进行扰动；仿真实验表明同,文中设计的基于改进量子粒子群算法和Map-Reduce模型能高效地执行任务,较传统的Map-Reduce模型具有较少的执行时间,具有很强的可行性,是一种有效的通用并行计算模型。

Abstract:

In order to realize effective management of user tasks in the large computer group, and conquer the defects of the low universality of the given parallel computing framework, a parallel computing framework is propoesd based on improved Quantum particle swarm algorithm and Map-Reduce model. Firstly, the classic Map-Reduce model and the parallel computing flow were described. Then the improved Map-Reduce model was designed based on improved Quantum particle swarm algorithm, the multi-population was parallel searched and the fitness was computed, and the particle was sorted and the particle population was divided, then the control coefficient and particle position were renewed in the Reduce stage, when the global solution was unchanged, the particle was changed by chaos interrupt. The simulation experiment shows the method in this paper can execute task effectively, and compared with the traditional Map-Reduce model it has the less execution time. Therefore, the method in this paper has strong feasibility and universal parallel computing model.