For details about the cluster and configuration of the glusterfs file system, see glusterfs practices.
6 servers
1. Single-client Testing
Test 1: copy a file of about GB to glusterfs (all small files, with an average of 1 MB)
Test results: glusterfs: 1.8 m/s. The throughput of the six servers is 3.6 m/s.
Test 2: read Test 1 files from glusterfs
Test result: glusterfs: 10.8 m/s local hard drive: 25.6 m/s
Test 3: Copy A 1g file to glusterfs
Test result: glusterfs: 2.334 m/s. Due to backup, the throughput is nearly 5 m/s.
Test 4: read large files copied in Test 2
Test result: glusterfs: 18.7 m/s
2. Multi-client Testing
Test 5: use five clients to read files copied from Test 1 to glusterfs
Test result: glusterfs: 25.7 m/s, which is about the Read efficiency of the local hard disk.
Test 6: 2 client small file write tests
Test result: glusterfs: 2.12 m/s
Iii. Summary:
The write speed of small files is only about 2 MB, and the speed is too low. Fortunately, the write speed is stable in the case of multiple clients. Writing large files is only about 2 MB. Because the sample database features one write and multiple reads, it is sufficient for the sample server.
In addition, in terms of performance tuning, You can optimize the configuration as needed, and there is still room for improvement in reading and writing.
The glusterfs Distributed File System features that multiple clients can use distributed resources more efficiently.
Network Data + test data show that glusterfs is relatively stable in terms of high availability. However, because updates in the replication mode are performed through the client, when the client and a server in the replicate are mounted at the same time, data will not be synchronized. You must manually create a list action (LS) to update the list.