Optimizing The Logging Mechanism Based On SSD In Ceph's Storage Engine

Ceph is a popular distributed storage system with excellent scalability and availability. With the development of materials technology, new storage devices, such as SSDs, become more and more popular. Many users of Ceph Cluster use the SSD to accelerate application access. In a typical deployment scenario, Ceph construct distributed cache pool by SSD, or use SSD as the log device of HDD on the cluster node to improve system performance. But Ceph's storage engine use the logging mechanism which need to update log and file on a write to implement transaction interface, causing "double write" problem and limiting the performance improvement by SSD.The hot data write delay mechanism which proposed to solve the "double write" problem of traditional logging mechanism when using SSD as a log device, equipment log into temporary storage space of hot data through memory index, thus recording the updates of hot data in log achieves the update of hot data; when recycling log space, the last valid update of hot data will be written back into file system, achieving postponed writes and I/O merged of hot data, thus reduce number of writes to disk; using double log structure to respectively record the update operation of hot data and cold data, and cold data is write directly to the file system as soon as appending to the log, reducing the reading overhead when recycling log space.The optimization of logging mechanism not only achieve atomicity of transactions, but also reduces the number of writes to disk by using write delay and double log structure. A number of performance tests show that the optimization can significantly improve system performance under fileserver load because of its significant hot data and can also provide the performance as original system under random write load because of its less hot data.