A QueryRating-Based Statistical Model for Predicting Concurrent Query Response Time

Zefeng Pei, Baoning Niu, Jinwen Zhang, Muhammad Amjad

Abstract: Predicting query response time plays an important role in managing database systems. It can be used for tasks such as query scheduling, resource allocation, and system capacity planning. Due to the uncertainty of database systems, especially query interactions between concurrent queries, query response time varies greatly from one run to another. At present, there are two types of models for predicting query response time, namely, analytical models and statistical models. Since these models do not quantify query interactions and select the modeling metrics properly, they are only suitable for predicting the response time of the medium size queries. To address the issues, this paper proposes to quantify query interactions using QueryRating, and introduce a new statistical model TMT&TP to predict the response time of queries without the size constraint. Experiments show that the average prediction accuracy of response time for any size of queries is as high as 83%.