The basic idea of the cache tiering is the separation of hot and cold data, the use of relatively fast/expensive storage devices such as SSD disks to form a pool as the cache layer, the backend with relatively slow/inexpensive devices to form a cold data storage pool.
The Ceph cache tiering Agent handles automatic migration of data from the cache layer and storage layer, transparently to client transparent operations. The Cahe layer has two typical modes of use:
1) Writeback mode
The Ceph client writes the data directly to the cache layer, returns immediately after writing, and the agent then migrates the data to the cold data pool in a timely manner. When the client takes cold data that is not in the cache layer, the agent is responsible for migrating the cold data to the cache layer. That is, the Ceph client makes IO Read and write operations directly on the cache layer and does not exchange data with a relatively slow cold data pool.
This mode is suitable for the operation of variable data, such as photo/video editing, e-commerce transaction data, etc.
2) Read-only mode
When the Ceph client writes directly to the back-end cold data pool while reading, Ceph reads the data from the backend to the cache layer.
This model is suitable for immutable data, such as Photos/videos on Weibo/video, DNA data, X-ray images, etc.
Crush algorithm is the core of Ceph cluster, on the basis of deep understanding of crush algorithm, utilize SSD's high performance, can use less cost to increase, meet the high requirement of storage performance for enterprise key application.
Please refer to the link http://docs.ceph.com/docs/master/rados/operations/cache-tiering/?highlight=tier for detailed procedure.
This article is from the "Attitude decides everything" blog, please make sure to keep this source http://sangh.blog.51cto.com/6892345/1627400
CEPH Cache tiering