一种元数据服务器集群的负载均衡算法

Load Balancing Algorithm in MetaData Server Cluster

在基于对象的存储系统中,元数据访问非常频繁,大规模存储系统中元数据的访问是潜在的系统性能瓶颈.元数据服务器集群中必须负载均衡,以防某个元数据服务器成为存储系统访问的瓶颈.现有文章中很少有研究元数据服务器集群的负载均衡的文章.本文中采用元数据请求的响应时间来衡量一个元数据服务器的负载情况,首先从映射算法上实现静态负载均衡,并针对元数据热度差别大而引起的负载不均衡引入动态负载均衡,通过仿真结果显示其有效性.

In object-based storage system, metadata is frequently accessed, which is the underlying storage system performance bottleneck. Load balancing must be implemented in metadata server cluster in order to avoid any metadata server becomes the bottleneck of the system. Few existing papers are about load balancing in metadata server cluster. This paper uses metadata request response time for measuring the load condition of a metadata server. A mapping algorithm is adopted at first to achieve the static load balancing, and the dynamic load balancing algorithm is introduced to solve the load imbalance brought by the difference of metadata popularity.