Performance of integrated concurrency control method in distributed database systems

Abstract

In this thesis, an integrated concurrency control (CC) method for fully replicated databases, called OPTLOCK, is presented, which allows both locking and optimistic scheme but adaptively selects the suitable one for each incoming transaction. The performance of OPTLOCK is also studied. To study the performance, a transaction processing model for fully replicated databases in which OPTLOCK can be handled is also presented, and the performance of OPTLOCK are compared with that of two-phase locking(2PL). OPTLOCK and 2PL are simulated under the same assumptions by using the concurrent language Path Pascal. Based on the simulation results, the average response time of OPTLOCK is found to be mainly affected by the read/write ratio. When the read/write ratio is high, OPTLOCK performs better than 2PL. However, when the read/write ratio is low, 2PL performs better than OPTLOCK due to the greater overhead in terms of the CC information processing time. When the read/write ratio varies, the rates of changes in the average response times in OPTLOCK are much greater than those in 2PL. In addition, it appears that the interarrival time and communication delay have some effect on the average response time. In general, when the interarrival time is large, i.e., the system load is light, OPTLOCK performs better than 2PL, while the reverse holds true when the system load is heavy. When the interarrival time varies, the rates of changes in the average response times in OPTLOCK are much greater than those in 2PL except the case where there are no transaction conflicts in the system. However, when the communication delay varies, the rates of changes in the average response times in both schemes are found to be similar.