Koozali.org: home of the SME Server
Contribs.org Forums => General Discussion => Topic started by: purvis on March 20, 2015, 11:24:43 PM
-
I had wrote some back on this subject back in October 2013.
http://forums.contribs.org/index.php/topic,50272.0.html
I am using SME 8.1
Today I am rebuilding an raid 1 array with 2 drives that are 2 TB in size.
1 of the 2 TB drives was having issues due to drive errors found also using S.M.A.R.T. drive diagnostics.
After putting in a new 2 TB drive like the one before(I keep drives at my location on the shelf for this purpose).
And starting the new build. The time was going to take over 447 minutes.
After setting the NCQ to 1, that time dropped down to about 190 minutes.( finish=185.6min speed=160835K/sec).
The report previous speed was about 71000K/sec or something close to that before setting NCQ to 1
I have moved pretty much moved other servers to all SSD drives and the times to complete a rebuild is relatively small in time compared to this 2TB standard hard drive.
I will not be checking back on this post.
This post was just meant to save other people some time.
smartctl returned this info
# smartctl --scan
/dev/sda -d scsi # /dev/sda, SCSI device
/dev/sdb -d scsi # /dev/sdb, SCSI device
The code I was able to use for setting NCQ to 1 was
echo 1 > /sys/block/sda/device/queue_depth
echo 1 > /sys/block/sdb/device/queue_depth
I will do the usual update with the below instructions to let the NCQ reset back to the default(likely 32) after the array rebuild process is complete
yum clean all
yum update -y
signal-event post-upgrade;signal-event reboot
-
I assume you mean "reducing rebuilding time".
-
Interesting, are there any contrary implications?
-
Reducing raid build time is right. I cannot edit the header to this post
-
I will not be revisiting this post.
But I do think it would be good to NCQ on raid 1 rebuilds to a factor of 1.
I have not checked any behavior with the raid check routine.
I have kill that function on our systems with SSD. I run it once in a great while manually.