Read Performance on ZFS Vol or FS ( dd test )

[mike7512]

Hi

i have a Kernel 3.0.4 with zfs 0.6.0-rc5 .

Write Performance on a raidz1 whit 3 x 2TB SATA Drivs is 240MB/sek
( dd if=/dev/zero of=/datapool bs=1M count=10000 )
( compress and dedup is Off )

Read performance is 70MB/sek ?
On zpool scrub i can see 300MB/sek Read ( iostat )

Any any ideas ?
Thanks for help !

[in4mer]

Was this superceded or duplicated by another ticket? I too have read bandwidth problems.

[ryao]

If I had to guess, I would say that read throughput was low because of various seeks that needed to be done on the disks for metadata (that is already cached when doing writes) and the iostat figure is the combined sum for IOs done on all 3 disks. In that case, this is not a bug.

With that said, @behlendorf likely closed this because it is hard to see what is actionable here.

[behlendorf]

Exactly right. I closed this issue because it was over a year old, we've fixed a ton of stuff in the code, and there wasn't enough detail for us to do anything.

[in4mer]

OK.

I have a specific use case that has unbelievably bad performance. Continuous reads from a single large media file, originally written to an otherwise quiescent array. The size of the file is roughly 8GB. The array, roughly 7TB.

If you take the average amount of data read from the array during playback, it totals roughly 2.3TB. No hyperbole.

Any objection to me opening a new issue about this?

[behlendorf]

@in4mer By all means open a new issue, and please provide as much detail about your setup as possible.