A New Look at Timestamp Ordering Concurrency Control
DEXA '01 Proceedings of the 12th International Conference on Database and Expert Systems Applications
Simulation models of two-phase locking of distributed transactions
CompSysTech '08 Proceedings of the 9th International Conference on Computer Systems and Technologies and Workshop for PhD Students in Computing
| Hi-index | 0.00 |
The increasing demand for high throughputs in transaction processing systems leads to high degrees of transaction concurrency and hence high data contention. The conventional dynamic two-phase locking (2PL) concurrency control (CC) technique causes system thrashing at high data contention levels, restricting transaction throughput. Optimistic concurrency control (OCC) is an alternative strategy, but OCC techniques suffer from wasted resources caused by repeated transaction restarts. We propose a new technique, ORDER, that enlists the aid of the interconnection network in a distributed database system in order to coordinate transactions. The network in an ORDER system provides total ordering of messages at a low cost, enabling efficient CC. We compare the performance of dynamic 2PL and ORDER, using both an analytical model and a simulation. Unlike previously-proposed models for 2PL, our analytical model predicts performance accurately even under high data contention. We study the effects of various parameters on performance, and demonstrate that ORDER outperforms dynamic 2PL for a wide range of workloads.