| The traditional approach of installing one operating system on one computer has been challenged over the last years by virtualization which utilized one physical machine to run in parallel multiple independent computers. The virtualization framework enabled several discrete computers to function simultaneously by using hypervisor software separating the operating system from hardware. While the cost savings of virtualization were compelling reasons for implementation, there were aspects hindering its adoption, such as complexity and lackluster performance. This study focused on the performance problems of virtualized databases and its participants, MySQL, PostgreSQL, and MariaDB, were a particular segment of software, known as open-source relational databases and defined as databases available free of license fees. The specific problem was the potential inferior performance of virtualized open-source relational databases, and the way such environments quantitatively compared to non-virtualized systems. The methodology used the quantitative study of relational database performance as measured by the computer parameters of processor, disk input output and memory utilization, and then compared such parameters in physical versus virtual environments. Another variable studied, the transactions-per-second, was a direct measurement of database performance. The methodology was practically implemented by using database benchmarks performed first in physical then virtual architectures and recording the processor, disk, memory usage and the transactions-per-second. Following virtualization, the database transactional capacity as measured by transactions-per-second was affected by as much as 14%. In the case of disk performance, the results, t(44) = 15.12, p < .01, suggested that the variable was considerably affected by virtualization. The statistical results for memory utilization, t(44) = 1.23, p = .11, have shown a non-significant statistical difference, indicating that the memory variable was not meaningfully affected by virtualization. Lastly, results for central processing unit, t(44) = 8.48, p < .01, suggested that the processor utilization variable was significantly affected by virtual migration. The overall conclusions drawn from the results of this study were that a transition of open-source relational databases from physical to virtual environments resulted in severe performance impact on disk input output and processor usage while the amount of utilized memory remained mostly unchanged. |
