No idea about MySQL. I use postgresql exclusively. And it screams pretty good on ZFS. You can set the record size. The sector size. If you’re insane you can mess with fsync. You can tune ZFS so easily to handle the workload you have.

> "Future posts will be devoted to more specific topics like performance, ..."

I'm looking forward to them myself.

Are there Facebook engineers here?

I believe Facebook heavily uses btrfs, which has a similar design and feature set. But they may only use it for object storage.

Did they try using it for MySQL databases?

I haven't done it on heavily enough MySQL servers to really tell you - but on "typical" installs there is no noticeable difference.

I'd say add a bit of extra RAM for ARC overhead you wouldn't usually, and of course invest in SLOG devices.

ZFS is never going to win any speed races, but what we gain in snapshotting/rollback/trivial remote backups makes it a great fit for many of our pgsql/mysql installs.

Conventional wisdom is that with basic (lz4) file system compression, the speed up you get in I/O r/w is substantial enough to be many tens-of-percentages performance improvement.

Correctly tuned, it could be faster on some workloads and hardware (block-level compression really saves on disk I/O).

Future articles should flesh that out.