Recurrent neural network-based database query time prediction method

Abstract

The invention discloses a recurrent neural network-based database query time prediction method. The method comprises the following steps of: firstly extracting query plans from database history query records so as to form original data, wherein each query plan comprises operation information and operation time; classifying the original data according to the lengths of the operation time to ensure that the quantities of the query plans in various types are the same; specially processing the query plans to obtain an operation sequence and an operation time sequence; taking the operation sequence as a feature vector, taking the operation time sequence as a label, inputting the feature vector and the label into a neural network, and carrying out training to obtain a model; for a to-be-predicted query plan, repeating the steps to obtain an operation sequence, inputting the operation sequence into the model, and outputting the operation time sequence to complete the prediction of a database query time. The method is capable of obtaining favorable effect in the aspect of relational database query time prediction, and the correctness of the model under simulated data training is higher than 78%. The method can be used for solving the key problems in query optimization and load management.

Description

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Login to view more

AI Technical Summary

Benefits of technology

It describes an improved method called predicting how fast queries are going on during real world scenarios like stock market prices overcomes previous methods based only on fixed data points such as price levels. By converting complicated operations into a single dimensionality, this approach improves performance compared to older models. Additionally, there're certain benefits from utilizing LSTM networks instead of traditional ones due to their ability to learn patterns quickly without being trained too hard. Overall, these improvements lead to faster and higher quality results when searching through large databases containing many different types of information.

Problems solved by technology

This patented technology addresses issues with estimating the number of times each SQL operation takes place on disk space (the size of memory needed). Current methods only consider one aspect - wait time prediction, while others use estimated costs instead of actual hardware usage. Additionally, current techniques involve expensive calculations involving multiple tables and complicated algorithms like matrix factorization.

Images