FreeBSD ZFS vs. TrueOS ZoF vs. DragonFlyBSD HAMMER2 vs. ZFS On Linux Benchmarks

[phoronix]

Phoronix: FreeBSD ZFS vs. TrueOS ZoF vs. DragonFlyBSD HAMMER2 vs. ZFS On Linux Benchmarks

With TrueOS offering daily snapshots built against the "ZFS on FreeBSD" code derived from OpenZFS / ZFS on Linux, I decided to run some benchmarks to see how the performance compares to that of FreeBSD 12.0 with its ZFS file-system support, DragonFlyBSD 5.2.1 with its HAMMER2 file-system alternative, and then Linux with ZFS/ZoL and other file-system options.

http://www.phoronix.com/vr.php?view=27459

[bitnick]

Michael Thanks for benchmarking ZFS!

However, I see ZFS more of a file system for cold data storage and redundancy, i.e. useful mainly for NAS:es and similar application. If I'm correct then maybe testing ZFS on a single SSD is missing the point a bit? I for one would be more interested in tests comparing ext4, XFS and ZFS on HDDs -- with ZFS running for example in a mirror configuration.

Of course, for synchronous database tests, ZFS should probably be configured with a ZIL cache on SSD? And ext4 should be running entirely on an SSD, maybe? XFS may not be the best filesystem for database backing at all?

Not so easy to set up, but the results would be a lot more meaningful.

[milkylainen]

Comment before I read. This will be as usual. HAMMER2 posting incredible because it's probably ignoring everything that remotely looks like a fsync.

[Raka555]

Comment before I read. This will be as usual. HAMMER2 posting incredible because it's probably ignoring everything that remotely looks like a fsync.

Wonder if HAMMER2 is giving any consistency guarantees ?

I would not mind if they optimize fsync's out as long as the filesystem is guaranteed not to be corrupt at any point if it loses power.

[aht0]

ask Matthew Dillon, he has user in Phoronix.

[hybridchemistry]

Michael Thanks for benchmarking ZFS!

However, I see ZFS more of a file system for cold data storage and redundancy, i.e. useful mainly for NAS:es and similar application. If I'm correct then maybe testing ZFS on a single SSD is missing the point a bit? I for one would be more interested in tests comparing ext4, XFS and ZFS on HDDs -- with ZFS running for example in a mirror configuration.

Of course, for synchronous database tests, ZFS should probably be configured with a ZIL cache on SSD? And ext4 should be running entirely on an SSD, maybe? XFS may not be the best filesystem for database backing at all?

Not so easy to set up, but the results would be a lot more meaningful.

+1 Thanks for benchmarking ZFS, I've been an avid user of it for sometime now and always interested in it's evolution.
I also agree with bitnick in that the tests with one single SSD, while interesting, don't seem to fit the use case of ZFS. Several ZFS arrays of HDDs against several mdadm, and or hardware RAID arrays would be a better benchmark IMO.

[Machine]

Was surprised by some results, although like the other commenters I think an array of disk test against the FSs would be interesting.

[Zyklon]

Michael Please add XFS next time!

[mbello]

Michael why is ZFS so much faster under FreeBSD? Should we expect ZFS performance to suffer now that FreeBSD is moving to ZoL's codebase?

[Zyklon]

Performance should rather get better, as ZoL is better maintained than the Solaris base, which FreeBSD used before.