Test and Optimization of Large-scale Ceph System

This article shares a test and optimization case based on a large-scale Ceph system, which covers functional testing, high availability testing and performance testing. In the process of deployment and testing, we find the OSD number confusion, the exorbitant initial cluster capacity, abnormal PG state of cluster, discrepancy of properties and expectations or other problems. We analyze the above problems, put forward corresponding solutions and verify. In the performance test, we carried out a theoretical estimation and stress test for the performance. The test results show that the large-scale Ceph system can run stably in production environment with high availability, and can show 50% increase high performance after proper optimization.