Identifying the Major Sources of Variance in Transaction Latencies: Towards More Predictable Databases

Decades of research have sought to improve transaction processing performance and scalability in database management systems (DBMSs). Far less attention has been dedicated to the predictability of performance—how often individual transactions exhibit execution latency far from the mean. Performance predictability is vital when transaction processing lies on the critical path of an interactive web service, or in emerging market segments that offer transaction processing as a service to customers who contract for guaranteed levels of performance. In this paper, we take several steps towards achieving more predictable database systems. First, we propose a profiling framework called VProfiler that, given the source code of a DBMS, is able to identify the dominant sources of variance in transaction latency. VProfiler works by deconstructing overall transaction latency variance into variances and covariances of the execution time of individual functions, which provide insight into the root causes of variance. Second, through a case study of MySQL, we show that lock management is a primary source of latency variance and propose a new lock scheduling algorithm, called VarianceAware Transaction Scheduling (VATS), to reduce variance. We additionally propose enhancements to the buffer pool replacement policy and identify MySQL parameters that can be tuned to reduce variance based on the output of VProfiler. Our experimental results show that our schemes reduce overall transaction latency variance by 37% on average (and up to 64%) without compromising throughput.