一个自组织可扩展PC集群存储系统

随着大型的分布式系统及商用机器集群的流行，许多重要的研究都致力于设计可扩展的集群存储系统。文中介绍了一个自组织可扩展PC集群存储系统的设计方案，并提出了一种对各个存储结点所挂磁盘的分组方案，对于大小数据对象采用不同的放置定位技术，重点讨论了小数据对象的伪随机放置算法，该算法在系统扩展时也能够有效地定位数据对象。该方案在同构系统中，达到了较好的磁盘负载平衡，从而各个节点的负载也达到了相对平衡的状态。另外通过对数据对象的有效的复制，使系统具有很好的容错性。

A Self-organized Scalable Cluster-based Storage System

As the prevalence of large distributed systems and clusters of commodity machines has grown, significant research has been devoted toward designing scalable distributed storage systems. This paper presents a self-organized scalable cluster-based storage system and a disk stripping scheme of the storage nodes of the system. A data object placement and location technique that differentiates between small and large data objects is introduced. The discussion is focused on the pseudo-random placement algorithm of small data objects, which can efficiently locate the data objects during the system scaling. Better disk workload balance can be acquired in the homogeneous system, consequently, relative workload balance status can be reached in various nodes of the system. In addition, system fault tolerance is achieved through the efficient replication of the data objects.