Data locality characterization of OLTP applications and its effects on cache performance

Online transaction processing (OLTP) is an important commercial application with intensive memory access. With increasing gap between processor and memory speeds, cache performance associated with data locality is the critical performance factor for OLTP applications. While OLTP applications have been shown to have significant data locality, data locality characterization and its effects on cache performance is still an open question. This paper presents a detailed study of typical OLTP queries from TPC-C. We identify a set of working sets by analysis of reuse in query processing. Based on performance experiments on typical OLTP queries with simulated system, we find cache performance is mainly impacted by critical working set that are caused by fundamental structures required to access tuples in relations. We also find that the critical working set is slightly affected by application parameters.