Approaches to Model Query Interactions

A typical database workload consists of a mix of multiple query instances of di erent query types that run concurrently and interact with each other. That query interactions can have a signi cant impact on database system performance. Hence, developing performance tuning algorithms for database systems requires taking these interactions into account. In this paper we propose three new approaches to model and measure query instances and query types interactions. Our approaches require no prior assumptions about the internal aspects of the database system or the reason of query interactions, making it non intrusive, namely, portable across systems. To demonstrate the pro t of exploiting query interactions, we have developed a novel interaction-aware query scheduler for online workloads. We report an experimental evaluation of the proposed approaches using TPC-H workloads running on PostgreSQL. The evaluation shows that our approaches have potential to improve many database tuning algorithms.