Page 6 of 10
Benchmark profile: Sysbench: - prepare: 100K inserted rows - simple: 32 threads / 100K read-only selects - complex: 32 threads / 10K read-write transactions pgbench: - scale: 100 - maxtrans: 100 - clients (threads): 32
We can do the same for PostgreSQL database.
Again, let use sysbench and populate the database:
BTRFS is very slow when populating a PostgreSQL database. XFS is better, but EXT3/4 filesystems are the best here.
Simple test now:
We see no real differences. Now, the complex test:
Now we see some big differences, as BTRFS is 2x slower then EXT4 in this case. EXT3 and XFS are also behind the leader.
Will pgbench confirm the same picture?
Only in part: while EXT4 remain the leader, now the slower filesystem is XFS. From the two databases benchmarked, it seems that you can not wrong using EXT4.