双均衡的集群存储资源映射方法

Dual-balance storage mapping method based on cluster storage

为了解决集群存储环境下的存储资源管理问题,提出一种双均衡的集群存储资源映射方法。该方法包含两个阶段:第一阶段基于LPT(longest processing time)算法求解集合划分问题,实现将虚拟存储资源请求均匀地分配到节点上;第二阶段基于Toyoda算法求解多维背包问题,用于进行节点内部设备级别的资源映射。这种两阶段的求解过程可以极大地简化集群存储资源映射问题的求解难度,并达到节点间负载均衡和节点内部多维度资源使用均衡的双均衡目标。模拟实验表明该方法不仅达到双均衡的资源映射目标,而且对不同维度、不同粒度的资源请求情况具有良好的适应性。

A dual-balance storage resource mapping method was developed to improve storage management problem in cluster storage. The method first uses the LPT （longest processing time） algorithm to solve a set partition problem to equally assign virtual storage resource requests to every node. Then, the second stage uses the Toyoda algorithm to solve a multi-dimensional bin packing problem to map the inner-node resources. This two-stage solution greatly simplifies the solution of the resource mapping problem for the cluster storage to achieve the dual balances of the load balance between nodes and the multi-dimensional resource utilization balance within each node. Simulations show that the method not only balances both targets, but also has better adaptability to different dimensions and sizes of resource requests.