遗传优化的混合网格计算调度模型SCE部署研究

Research on SCE Deployment of Hybrid Grid Computing Scheduling Model Based on Genetic Optimization

网格计算涉及的资源通常存在区域和组织差异,涉及的作业(Job)则存在数据和计算两种密集类,针对具有混合特征的网格计算,提出了基于SCE中间件的遗传优化网格作业调度算法。首先分析了具有不同密集类型Job的混合网格计算模型,得到作业与资源的属性表示,以及作业调度与资源之间的约束关系。然后根据混合网格计算模型分析,将其转化成约束条件下的最优解问题,引入改进遗传算法进行求解。在种群初始化时根据适应性筛选出一部分样本作为初始种群,利用高质量样本启发寻优,降低进化代数;同时针对每个染色体的作业执行速度和染色体内每个作业的执行速度依次设计适应性,从而加速收敛;通过适应性修正、交叉和变异处理,防止种群出现过早或者局部收敛,并且增加种群多样性。最后基于SCE部署作业调度,从中间件进一步提升作业调度效率,减少出错。实验结果表明,基于SCE中间件的遗传优化网格作业调度算法能够有效抑制执行错误的发生,提升作业调度与资源配置的效率,降低作业调度响应时间。

The resources involved in grid computing usually have regional and organizational differences, while the jobs involved have two intensive categories of data and computing. For grid computing with hybrid characteristics, a genetic optimization grid job scheduling algorithm based on SCE middleware was proposed. First, the hybrid grid computing model with different intensive types of jobs was analyzed, and the attribute representation of jobs and resources, and the constraint relationship between job scheduling and resources were obtained. Then, according to the analysis of the hybrid grid computing model, it was transformed into the optimal solution problem under the constraint conditions, and the improved genetic algorithm was introduced to solve the problem. During population initialization, a part of samples were selected as the initial population according to the adaptability, and the high-quality samples were used to inspire optimization and reduce the evolutionary algebra. At the same time, the adaptability was designed according to the execution speed of each chromosome and the execution speed of each job in the chromosome, so as to accelerate the convergence. Through adaptive modification, crossover and mutation, premature or local convergence would be prevented and population diversity would be increased. Finally, based on SCE deployment job scheduling, the efficiency of job scheduling was improved from middleware and errors were reduced. The experimental results show that the genetic optimization grid job scheduling algorithm based on SCE middleware can effectively suppress the occurrence of execution errors, Improve the efficiency of job scheduling and resource allocation, and reduce the response time of job scheduling.