基于Chord算法的可扩展高性能元数据存储环设计

Design of Scalable High-performance Metadata Storage Ring Based on Chord Algorithm

Hazel Nut是一个基于块的分级存储系统,为满足该系统元数据内存需求量大、访问速度快、动态可增长的要求,设计并实现一种可扩展高性能的元数据存储环SCRing。SCRing由元数据存储环shell ring、映射环chord ring和缓存表组成,shell ring与chord ring分别用于存储与定位数据块元数据信息,缓存表用于缓存数据块元数据在内存中的位置。通过消息摘要算法计算得到SCRing双环节点间的映射关系。依据SCRing对消息摘要算法的实际需求,结合Linux内核版本,测试并比较Linux内核支持的消息摘要算法的速度、均匀性和稳定性,得出SCRing所应采用的算法种类和使用模式,并在此基础上对SCRing读写性能进行验证,结果表明SCRing具备可扩展性和高性能,满足Hazel Nut系统元数据访问需求。

HazelNut is a block-based hierarchical storage system, which has special metadata requirements, including huge memory cost, fast access speed and everlasting data growth. To meet those requirements, a scalable high-performance metadata storage ring,named SCRing,is presented. SCRing consists of double rings and a cache table. One of the rings is named as shell ring to store the metadata information of data block, and the other is named as chord ring to locate the metadata information of data block. The cache table is used to cache the memory address of block metadata. The mapping relationship between the two rings are got through message digest algorithms. According to the SCRing demands on message digest algorithms and the Linux kernel version, a series of tests is made on digest generation speed, uniformity and constancy to get algorithm category and usage mode. Besides,I/O performance test is done on the basis of the above message digest algorithm tests. The results prove that SCRing is featured with high performance and scalability, which meets the metadata access requirements of HazelNut system.