基于网络容量限制的分布式数据库的数据迁移

摘要

In the distributed database system, imbalance of load ofen occurs. Data migration is executed to rebalance load. Distributed database system faces challenges for data migration:the migration cost, say migration time and network interference. This paper focuses on how to minimize the cost while executing data migration. Previous research takes migration time as the migration cost, but does not take network interference into consideration. A new migration cost model is proposed in this paper which adds network interference into consideration. This paper designs a cost-aware algorithm based on network capacity matrix which aims to minimize the migration cost, as while as network interference. An experiment is done in order to prove the effectiveness of the algorithm. The experiment demonstrates that the model and algorithm proposed in this paper can rebalance load and reduce network interference compared with other models.%分布式数据库中数据时常会产生数据倾斜的现象，为了平衡数据库的负载，需要对分布式数据库进行数据迁移。分布式数据库的数据迁移成本主要包括两个方面：迁移时间和对网络性能的影响。论文主要解决在数据迁移过程中如何减少迁移成本的问题。论文研究了C Lim等人的数据迁移成本的模型，这些模型只把数据迁移的时间作为数据迁移的成本，并没有考虑到数据迁移过程中对网络性能的影响。论文提出了一种基于网络容量限制的数据迁移模型，该模型将数据迁移代价的因素中加入网络容量的限制，并提出了一种基于网络容量矩阵的数据迁移算法，实验证明该算法能够有效平衡数据倾斜，与最短迁移时间模型相比，该模型能减少数据迁移对系统网络性能的影响。