Lower than expected performance

[4ever6]

I've recently converted 8x6TB disks into 4 mirrored vdev's to play with ZFS (1.6.1). The performance is much less than expected, particularly with writes--like, 40-50MB/s. That's much less than the speed of a single disk, so something seems odd. No compression or dedup enabled, default 128k record size. I'm on an iMac using 2x Thunderbolt enclosures that were previously doing work w/ SoftRAID.

Before I just spray a fire hose of output - is there any particular command(s) that would be helpful?

[Brendon]

I'm guessing that that you have used the wrong ashift value for your disks. Should use ashift=12 for 4k block disks. Makes a big difference.

Brendon

[4ever6]

I'm guessing that that you have used the wrong ashift value for your disks. Should use ashift=12 for 4k block disks. Makes a big difference.

Brendon

I wish that were the case. ashift=12 in my pool, and all 8 drives are 4k physical.

[haer22]

Do you have a cache and/or log device? Can make a big difference
What is the load? writing big files? a lot of small? server?
What is the settings of sync and logbias? I played around with setting sync=disabled and it made a huge difference. Dangerous but was OK in my environment.
Are the disks in one chassi or several? Some chassis have a maximum throughput per chassi and that may hit you. If you have several chassis, put the mirrored disk on separat chassis.
How much memory does you host have? How much is available to zfs? check the output of "arcstat.pl 15". The two last figures show how much memory zfs has for itself.

I have 2 large pools (5*8TB raidz2, 7*4TB raidz2) working as a server. Access is only via network so the bottleneck is usually the network. zfs is not a speed daemon but it is safe

[Brendon]

Any progress?

-- Brendon