Improving Performance of Database Appliances on Distributed Object Storage

Abstract

Distributed object storage system has advantages in scalability and accessibility over standard block and file storage. However, the object approach lacks maturity when it comes to real-time systems such as transactional databases that are constantly being written. Specifically, the object approach cannot rival block-based systems for the dynamic read and write speeds required by disk resource-intensive applications such as databases. As more organizations migrate to cloud hosting solution, the need to address rapid application read and write will become top of the requirements list. This paper proposes a method to increase I/O performance of database appliances running in the cloud environment with distributed object storage as the underlying data stores. The proposed method involves separating the distributed storage's journal and data partitions to different hard drives and also separating a few database application directories to multiple RBD images from different storage pools in order to speed up the I/O operations. Experiments with SATA, SAS, and SSD type-drives with Ceph distributed storage system have been conducted based on proposed method and the results show significant performance compared to local drives and default distributed storage setup.