Performance modeling of distributed timestamp ordering: Perfect and imperfect clocks

Abstract

This work presents a model of a distributed database system which provides the framework to study the performance of timestamp ordering concurrency control. Locking and timestamping are two popular approaches to concurrency control in database systems. Timestamp-based algorithms have been proposed to protect distributed databases from inconsistencies during concurrent access. In these algorithms, transactions may reach a particular site in different order than their timestamps, due to unexpected network delays. This causes conflicts which the distributed concurrency control mechanism has to cope with. We exhibit an analytical solution, which has been tested with extensive simulation. The accuracy seems to be very high. We assume perfect and also imperfect clocks for synchronization and quantify the way in which local clock inaccuracies affect the phenomenon of transaction conflicts. In particular, we derive a lot of interesting performance measures such as probability of abort, mean waiting time, throughput, mean queue length and others.