Database Query Cost Prediction Using Recurrent Neural Network

Query cost models are the key parts of relational database workload management and performance tuning.Firstly,it is difficult,even impossible,to precisely estimate the costs of different relational operators,due to the complexity of database systems and competition of computer resources.Secondly,most previous research work does not address the problem of predicting actual execution time of a query and predicts the query performance by the cost like query optimizers generate.Thirdly,most existing research work uses general query information,while it does not take advantage of actual operators because of the complexity of query plans.To reduce the complexity of workload management,in this paper,we propose an elaborate cost prediction model based on recurrent neural network,which learns from operator behavior and detailed runtime information.In particular,the model uses a special kind of recurrent neural network,called Long-Short Term Memory(LSTM).Given an ad-hoc query,the model is able to predict its running time range before it starts to run.The model,which is trained continuously,predicts query cost dynamically for the running database instance.When it comes to the cost prediction of complex queries,traditional database cost models are not accurate because of correlation.However,our model has learned correlation by the neural network.Thus,it is more accurate in cost prediction.Our research proposes that novel approach to solve the key problem in database workload management.Verified by the experiments,the accuracy of the model is over 71%which shows the method is feasible to some degree.