[mythtv-users] Raid configurations
Terrence Martin
tmartin at physics.ucsd.edu
Fri Aug 13 21:54:39 EDT 2004
Shawn wrote:
>Anyone see any performance problems if I used four WD2500JD drives hung
>off a Promise SATA150 TX4 in software RAID6 for video storage? Of
>course, XFS on top of that.
>
>I just really don't want a single failure on a huge RAID0 video volume
>to destroy ALL my video.
>
>I figure that'd satisfy my storage cravings for a while, and be
>virtually free of worry for me, barring some form of corruption.
>
>
>
>
>------------------------------------------------------------------------
>
>_______________________________________________
>mythtv-users mailing list
>mythtv-users at mythtv.org
>http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
>
>
Far from having performance issues you will find performance to be very
nice indeed.
I run this for my storage array at home. Here is my config.
FC2 + Promise 150TX4 SATA controller + 4x Seagate 160GB SATA Drives +
PIII 800Mhz + 2.6 Kernel + XFS
I just ran a test.
52MB/s read, 33MB/s write
Definitely not performance to sneeze at for a $60 SATA controller and a
3 year old computer. Particular on writes since raid5 is not know for
its write speed. The Linux kernel has a very fast RAID5 implementation.
My tests are done with a simple dd command
#Write test
time dd if=/dev/zero of=bigfile.zero count=1000000 bs=1024
sync
# Read test
time dd if=/dev/bigfile.zero of=/dev/null
Then I do some quick bc math.
I think this is a pretty good test for myth since this is similar to
what you expect for video, long multi gigabyte reads and writes. My
tests are 1GB which is approx 3 times RAM so disk cache effects would be
minimal.
33MB/s will give you plenty of head room for multiple tuners.
A word of caution. RAID5 is the most disk efficient of the RAID
personalities. However it is also more risky that a true data mirror.
There is a non-zero chance that when you go to rebuild your array after
a disk failure that one of your other disks could fail during that very
intensive and time consuming process. I have first hand experience with
this sort of thing on a much more expensive hardware array (4.8TB
gone...:( ). If you can when a drive fails copy the data onto another
disk or some other backup. Then rebuild the array with the new drive.
Better to be safe than sorry.
Also if you have a unclean shutdown the array will take several hours to
re-initialize itself. During this time you will get reduced performance
during this initialization phase, although linux throttles this process
to 10MB/s. This is different than a fsck (ext2) and file system type
does not effect it. It is simply a requirement in raid5 on unclean
shutdown to scan every block and test the parity.
Those caveats aside I have run RAID5 for many years both at work and at
home and it is a good compromise between data security, cost and
performance. I have lost 3 disks over the last 3 years but not one bit
of data.
On the disks, it is less space but if you get 4x200GB seagates ($131 ea
at newegg.com) you get 5 years of warranty. You may only get 1 with the
western digital. Also the Seagate are native SATA drives. We have pretty
much switched to seagate for all of our cluster nodes where possible.
Too much hassle with WD and Maxtor.
Terrence
More information about the mythtv-users
mailing list