基于多目标优化的复本分布及任务调度算法

Replica location and task scheduler algorithm based on multi-target optimizing model

为了提高分布式并行服务器的系统效率,并保证数据可用性,提出了基于多目标优化的复本分布及任务调度算法.文章针对分布式并行服务器的特点研究了多目标优化模型,并以此模型为基础设计了复本分布策略和相应的任务调度策略.系统根据任务需求周期性地调整数据复本数量和驻留节点.它不仅保证了数据的可用性要求,而且大幅度减少了所有任务耗费的系统资源.分析和测试表明:在相同条件下完成相同任务,该算法耗费的系统资源小于常用的三模冗余.

In order to enhance efficiency of a distributed parallel server and ensure availability of data in the system, a replica location and task schedule algorithm is proposed, that is a multi-target optimizing model. The system periodically adjusts the redundancy and location of replica. The algorithm not only ensures availability of data, but also greatly reduces the quantity of system resources that all tasks require. By analyzsis and testing, it is shown that the cost of system resources applied to the algorithm is far less than 3-redundancy, when finishing the same task in the same condition.