Research On Performance Optimization For InfluxDB Time Series Database

As a new type of database in the database field,the InfluxDB time series database is widely used in scenarios such as the Internet of Things,the Internet of Vehicles,and the Industrial Internet.As a key factor affecting database performance,configuration tuning has also become a research hotspot in the database field.However,most of the current database configuration and tuning work is still done by experienced database administrators(Database Administrators,DBAs).Manual parameter tuning based on DBA experience can no longer guarantee to recommend a configuration combination with optimal or near-optimal performance within a limited time.Therefore,in order to further improve the accuracy of parameter configuration tuning and optimize the overall performance,it is urgent to study the intelligent tuning technology of InfluxDB parameter configuration.In response to the above problems,thesis designs and implements an automatic tuning framework for InfluxDB parameter configuration based on deep reinforcement learning,and on this basis,proposes a time-triggered mechanism to ensure timely alarm tuning of the model.In terms of tuning the model,a sample pool classification algorithm is proposed,which classifies and stores the sample data according to the size of the reward value,and optimizes the storage scheme of the training data.Secondly,an effective reward function is also designed to guide the model training.This improves the accuracy of tuning model.In terms of time-triggered mechanism,study the time series forecasting model,optimize the input and output dimensions of the forecasting model,encode the tuning time as input data,so as to trigger the tuning model in due course,thereby improving the overall performance of the system.Finally,thesis conducts experiments on the three proposed improvements based on deep reinforcement learning algorithms and time series forecasting models.The test results show that compared with the default configuration,the automatic tuning framework implemented in thesis can recommend a better parameter configuration combination in time,which increases the query throughput by about 25%,and reduces the query execution time and average response time by about 15%.,the memory usage of batch writing is reduced by about 10% to 20%.