基于HBase的地理分布副本管理机制

发布时间：2018-07-10 15:35

本文选题：分布式系统 + 一致性　；参考：《计算机应用》2015年11期

【摘要】：针对分布式存储系统中数据通常在多个数据中心有冗余的副本进行备份,需要健壮的机制维护各个副本的一致性,对分布式系统的副本复制理论作了深入研究后,提出了一套管理地理分布副本的算法。微软研究院提出服务等级协议,把用户对一致性的要求分成若干级别,每个级别与用户可容忍的延迟有关。系统保证在可容忍的延迟范围内,用户能拥有较高的服务等级。Tuba系统拓展了Pileus,允许系统根据所有用户发送的统计信息动态地改变主从副本存放的位置,以提高系统的平均性能,但Tuba系统的复制只是基于单个目标单位进行。对Tuba系统中的方法作出改进,提出了一套改变主从副本存放位置的算法,并在HBase分布式系统的副本复制中实现了该机制。系统完成后,通过实验验证了在改变主从副本存放位置时综合考虑两个region的相关性可以提高系统整体的效用。
[Abstract]:In the distributed storage system, there are redundant replicas in many data centers, which need a robust mechanism to maintain the consistency of each replica, so the replica replication theory of the distributed system is deeply studied. A set of algorithms for managing geographical distribution copies is proposed. Microsoft Research proposes a service level agreement that divides the user's requirements for consistency into several levels, each related to the user's tolerable delay. The system guarantees that within the tolerable delay range, the user can have a higher level of service. Tuba system extends the Pileus. it allows the system to dynamically change the location of the master and slave copies to improve the average performance of the system based on the statistics sent by all users. But the replication of the Tuba system is based on a single target unit. This paper improves the method of Tuba system, proposes a set of algorithms to change the location of master and slave copies, and implements this mechanism in replica replication of HBASE distributed system. After the completion of the system, it is proved by experiments that considering the correlation of two region comprehensively when changing the location of master and slave copy storage can improve the effectiveness of the system as a whole.
【作者单位】：中山大学数据科学与计算机学院;
【基金】：国家自然科学基金资助项目(61379157)
【分类号】：TP333

【相似文献】