Cache consistency and concurrency control in a client/server DBMS architecture

Abstract

This paper examines five application cache consistency algorithms in a client/server database system: two-phase locking, certification, callback locking, no-wait locking, and no-wait locking with notification. A simulator was developed to compare the average transaction response time and server throughput for these algorithms under different workloads and system configurations. Two-phase locking and callback locking dominate no-wait locking and no-wait locking with notification when the server or the network is a bottleneck. Callback locking is better than two-phase locking when the inter-transaction locality is high or when intertransaction locality is medium and the probability of object update is low. When there is no network delay and the server is very fast, no-wait locking with notification and callback locking dominate two-phase and no-wait locking.