Leveraging Lock Contention to Improve OLTP Application Performance

Cong Yan, Alvin Cheung

Proceedings of VLDB 2016


Locking is one of the predominant costs in transaction processing. While much work has focused on designing efficient concurrency control mechanisms, not much has been done on understanding how transaction applications issue queries and leveraging application semantics to improve application performance. This paper presents QURO, a query-aware compiler that automatically reorders queries in transaction code to improve performance. Observing that certain queries within a transaction are more contentious than oth- ers as they require locking the same tuples as other concurrently executing transactions, QURO automatically changes the application such that contentious queries are issued as late as possible. We have evaluated QURO on various transaction benchmarks, and our results show that QURO-generated implementations can increase transaction throughput by up to 6.53x, while reduce transaction latency by up to 85%.

[Project website]