Review on Automatic Virtual Machine Configuration for Database Workload

A Review on the the Article : Automatic Virtual Machine Configuration for Database Workoads

Mark Allan Paul Y. Sebastian July 9, 2010

1

Summary

Virtual Machine Monitors(VMM) are becoming popular tools for deploying Database systems and other software systems. With the help of the VMMs, virtualization is achieved and this allows a programmable and flexible layer of software between the application software and the physical resources. With this is mind, the goal of the authors of the article is to create VMs and automatically configure them on which the database management systems run and compete against each other for the resources one physical machine. They first showed a motivating example. They created 2 identical virtual machines using Xen VMs. They ran PostgreSQL on one and DB2 on the other and the VM instances ran on the same physical server. They ran workloads on each DBS, 1 instance of TPC-H query Q17 on a 10GB database on PostgreSQL and 10GB TPC-H database consisting of 1 instance of TPC-H Q18 on DB2. Also 50% of the CPU and memory capacity were allocated for each instance. The allocation of resources were equal but after the application of the configuration technique it recommended 15% of the CPU and 20% of the memory be allocated for the VM that ran PostgreSQL and 85% of the CPU and 80% of the memory be allocated for the VM that ran DB2. The results as seen on Figure 1 show a slight degradation of the performance of the PostgreSQL VM(7%) and a great improvement in the DB2 VM(55%). The reason is because the PostgreSQL workload is very I/O intensive in the execution environment, meaning it is not affected by the changes in the CPU or memory allocation. It is the inverse of the DB2 workload environment because it is CPU intensive which means it is sensitive in the changes of CPU or memory allocation. To make this possible, the information about the anticipated workload of each DBMS is needed to determine the appropriate configuration for each VM which the...