按序选取的并行存储系统数据分布方式研究

针对采用轮转方式的并行存储系统在增加新存储节点后的重均衡操作中移动数据量大、资源消耗高的问题,提出了按序选取的数据分布方式.它通过多次模拟数据重均衡操作来完成并行存储系统的数据分布,从而使得系统更易于节点扩展.基于此,又提出了优化按序选取方式,这样可以通过改变数据抽取规则来提高按序选取方式的访问并行度.理论分析和模拟实验表明,采用所提方式的并行存储系统,数据重均衡操作均可达到零无效移动率,从而最大程度地减少数据的移动量,降低系统的资源耗费.

Data Layouts of Parallel Storage System Based on Ordinal Selection

Aiming at the problem that when new storage nodes are added to a parallel storage system by using Round-Robin method,the rebalancing operation needs to move a large quantity of data among storage nodes and therefore consume plenty of system resource,a new data layout called ordinal-selection was presented.The data layout of parallel storage system was completed by using rebalancing operations of simulation data repetitiously so as to make the nodes extension of the system more easily.Furthermore,an optimized ordinal-selection layout was also presented to increase the system's parallel degree by changing the data selection rules.The theoretical analysis and experimental results show that for a parallel storage system employing one of these two data layouts,the rebalancing operation can achieve zero ineffectual-move-ratio,therefore the remove of the data can be decreased farthest and the system resource consumption is decreased.