Is data contention slowing down your database? This study tackles the performance of dynamic locking in centralized databases, employing an analytic model to understand the impact of data and resource contention. The analysis uses steady-state average values, resulting in a cubic equation with one valid root within the relevant parametric range. The model's predictions show strong agreement with simulation results for transactions requiring up to twenty locks. It effectively isolates data contention from resource contention, enabling a separate evaluation of their impacts and interactions. The study also demonstrates that systems with non-uniform access or shared locks are equivalent to systems with uniform access and exclusive locks. Ultimately, the research identifies an upper bound on transaction throughput imposed by conflict-induced blocking, leading to a rule of thumb for acceptable data contention levels. It further suggests strategies such as transaction restarts or lock pre-declaration to exceed this bound, with considerations for restart costs and resource contention. These insights provide practical guidance for optimizing database performance.
Appearing in ACM Transactions on Database Systems, this paper directly addresses the journal's core interest in database management and performance optimization. The study's analysis of locking mechanisms and their impact on transaction throughput contributes to the broader discussion on concurrency control, a fundamental topic within the database systems research community. Its findings are relevant to database designers and administrators seeking to enhance system efficiency.