Performance comparison of resilient concurrency control algorithms for distributed databases

Abstract

In locking-based concurrency control algorithms for distributed databases, two basic strategies are usually used to control the allocation and deallocation of locks to transactions: centralized locking (CL) and distributed locking (DL). Recently, there has been some debate on which strategy is better. Although previous work has shown that in a failure-free environment, the CL algorithm performs better than the DL version, there are still some doubts about the performance of the CL algorithm in an environment where failures could occur. Thus, in this paper, the performance of a resilient CL algorithm is compared with that of a resilient DL algorithm in an environment where site failures could occur. The results show that the resilient CL algorithm still outperforms the resilient DL algorithm in terms of mean response time of transactions, resource utilization and communication cost. In addition, it is shown that the reliability of the resilient CL algorithm is improved considerably and is comparable to that of the resilient distributed version when an election protocol is used to rapidly elect a new central site whenever the central site fails.