Many operators have selected OpenStack as their control plane of choice for providing both internal and external IT services. Year after year, the OpenStack user survey has shown Ceph to be the dominant backend for providing persistent storage volumes through OpenStack Cinder. As developers build new applications, and repatriate old workloads, they are quickly discovering the need to provide database services within their OpenStack infrastructure. Given the ubiquity of MySQL, and it’s reliance on persistent storage, it is of utmost importance to understand how to achieve the performance demanded by today's applications. Databases like MySQL can be incredibly IO intensive, and Ceph offers a great opportunity to go beyond the limitations presented by a single scale-up system. Since Ceph provides a mutable object store with atomic operations, could MySQL store InnoDB pages directly in Ceph?
This talk will first review the general architecture of Ceph and then we'll discuss the results of a series of benchmarks done on small to mid-size Ceph clusters. These benchmarks have lead to the development of prescriptive guidance around tuning Ceph storage nodes (OSDs), the impact the amount of physical memory, and the presence of SSDs, high-speed networks or RAID controllers.