Optane SSD RAID Performance With ZFS On Linux, EXT4, XFS, Btrfs, F2FS

[ryao]

As said in the article, Bcachefs is coming in its own article next week. It won't be part of any regular test articles (due to the time involved, etc) until it's 1. mainlined 2. proved its usefulness.

If you ever benchmark ZFS again, I suggest running `zfs recordsize=4K $POOLNAME` before creating any files. You will see an amazing performance boost in your benchmarks, since that would optimize ZFS for random IO, as opposed to the stock configuration that is optimized for sequential IO. You can basically ignore tuning guides and just run that configuration alongside the stock one to get a full view of what ZFS can do.

Also, if you are benchmarking with two disks, don't use raidz. That is like running RAID 5 with 2 disks, which is pointless. You want `zpool create $poolname mirror $disk0 $disk1` for a RAID 1 equivalent or `zpool create $poolname $disk0 $disk1` for a RAID 0 equivalent.

Lastly, your application start time benchmarks favor ZFS because its cache isn't being properly cleared during your benchmark runs. You might want to restart between runs to ensure that cache is cleared.

[deusexmachina]

Everyone complains that ZFS is not properly optimized, but let's see a single reputable benchmark from someone who did optimize ZFS properly and shows it performing well against the rest on 1 NVMe or mirrored NVMes. Sucks that we can't have data integrity and performance without introducing a lot of complexity - if only ZFS were faster!