RAID Performance?

Hi All,

I've just come to try and establish roughly what level of read and write performance I should expect from a 5 disk RAID 6 array using software raid (MDADM, EXT4 filesystem) within openmediavault

I used DD to create a 10GB tempfile on my setup and thought that the performance was awfully poor.... I achieved a 77.2MB/s write speed and a 110 MB/s read speed.... is this roughly what you would expect?

My experience of RAID arrays tells me something isn't quite right here. It's not throttled by CPU as I have a powerful enough CPU in there to handle transcoding with Plex (It's an i5 quad core running @ 2.8Ghz)

Any idea's on what I may have done wrong, or steps I can take to improve the performance. I'm not even fully saturating a gigabit link.

If you need any more info just let me know...

Cheers
Danny

Yes, array is active and rebuilt.

root@nas:/media/b336b873-849f-497c-bfe3-c14a78d83441/Restore# cat /proc/mdstat
Personalities : [linear] [multipath] [raid0] [raid1] [raid6] [raid5] [raid4] [raid10]
md0 : active raid6 sdb[0] sdf[4] sde[3] sdd[2] sdc[1]
      8790795264 blocks super 1.2 level 6, 512k chunk, algorithm 2 [5/5] [UUUUU]


unused devices: <none>

I expected a write penalty... double parity is obviously an overhead. But the impact to the read performance surprises me.
The computer is not exactly a slouch and as I mentioned I do not see a CPU bottleneck, which I would expect if the parity was causing the issues.

Now I have done RAID before mainly using motherboard softraids... and from my experience with RAID 5 (softraids generally dont do RAID 6) never had much of a read performance hit over a plain old two disk stripe but there was a performance penalty with writing. From this I kind of assumed a similar concept would apply to RAID 6 in terms of read speed. I'd expect a 5 disk RAID 6 array to be roughly on par with a 3 disk stripe for reading only

I did, however the latest benchmarks I found were from 2011. Even they were around the 120MB/s write and I couldn't see any for read.

I'm wondering if my choice of filesystem could have had an impact.... maybe I should have chosen XFS?

Also would using LVM have made any difference - I'm not sure what it offers me and I now have everything built without it so can't play!!

OK, thanks for your help but I think I've managed to find the answer...

http://askubuntu.com/questions…-mdadm-raid-6-write-speed

Basically mdadm appears to use a default stripe_cache_size of 256 - I have changed mine to 8192 as the first post suggested and instantly got write speed to 324 MB/s...

I will play with the figure a bit more and see what I can get out of it I think.

I agree with the comment on that page... I think the default needs revising!!!

Thanks again
Danny

Thanks for pointing me to that thread... I'm glad you told me about /etc/rc.local otherwise it would have probably caught me out.

Whilst I wouldn't say I'm brand new to linux it certainly isn't my strong point so even though I might have finally figured out that setting wasn't persisting a reboot without a pointer I wouldn't have known how to make the command execute automatically at startup.

Cheers
Danny