A pipeline-based approach for long transaction processing in web service environments

Abstract

In web service environments, long transactions need to lock resources – often database services – for a long time during their long execution duration. This would bring down the performance of transaction processing systems. The transaction compensation is a feasible solution through allowing sub-transactions to independently commit, however, it is not able to speed up the transaction processing. This paper proposes a novel pipeline-based transaction processing (PLbTP) model for Serial Long Transactions (SLTs), which parallelises the transaction processing to reduce the transaction execution duration. Furthermore, we design a time-stamp-based deadlock prevention mechanism for the control of multiple concurrent transactions. The simulation results demonstrate that our approach can significantly improve performance of SLTs without the aid of compensating transactions.