A CBR Model for Workload Characterization in Autonomic Database Management System

Abstract

For effective workload management and performance tuning in Database Management System (DBMS) the Database Administrators (DBAs) have to deal with many issues. Workload monitoring and controlling can make the things easy for a DBA. Workload type prediction and adaptation can enable monitoring and controlling of workload that helps in DBMS performance tuning. In this study we propose a Case-Based Reasoning (CBR) model for workload type prediction that also has the ability to adapt dynamic workload behavior. To observe the accuracy, effectiveness, significance and adaptiveness of the proposed CBR model, it is compared with existing well-known machine learning approaches, such as, Support Vector Machine (SVM) and Neural Network (NN). For the validation of the proposed CBR model many standard benchmark workloads are experimented using the MySQL DBMS. The standard TPC-C and TPC-H like queries are used for generating training and testing data. In this study various experiments have been performed for Online Transaction Processing (OLTP) and Decision Support System (DSS) workloads. The proposed CBR model characterizes the workload through predicting its types. At the end, for result validation we have performed post-hoc tests which shows that the proposed CBR model produces better results.