Instead, you should run “/sbin/fstrim -a” once a day, in a cron job.
An actual optimization for ext4 is “noatime”, if you’re sure you don’t need up to date access times on your files. “nobarrier” might not be worth the risk, but to see that you need to compare setups that differ in only one setting.
Wonder if omitting logbias=throughput on the data files filesystem was intentional, or if maybe zfs-on-linux doesn’t support it?
That’s just FUD, they work perfectly fine with SSD.
You do not want want to TRIM after each and every operation, because it completely kills your performance: http://people.redhat.com/lczerner/discard/ext4_discard.html
Instead, you should run “/sbin/fstrim -a” once a day, in a cron job.
An actual optimization for ext4 is “noatime”, if you’re sure you don’t need up to date access times on your files. “nobarrier” might not be worth the risk, but to see that you need to compare setups that differ in only one setting.
Unless you write more than the volume of the disk every day.
Anyone with ZFS knowledge have any idea if the values they got are reasonable? Or is it slow because zfs on linux is immature? Poorly tuned?