4 disk RAID5 running too slow - pls help!

dan_aka_jack

dan_aka_jack

dan_aka_jack

Associate
Joined
8 Mar 2004
Posts
409
Location
London, UK
Hi there,

I have a 4 x 500GB RAID5 array using the Highpoint RocketRAID 2300 PCI-E controller. HDTach reports that I'm only getting 65MB/sec average read speeds which is much slower than others are getting. I've tried turning off NCQ but that only bought me a few MB/sec.

Any clues?!

Specs:

AMD Athlon 64 X2 2.4GHz
3GB RAM
WinXP Pro SP2
Latest Highpoint drivers
nVidia nForce4 mobo

Thanks,
Jack
 
I would certainly be expecting more from a 4 disk array than that, probably 150Mb/s at least. Remind me which disks you went with, WD AAKS's?

Can you post a screen shot of the HDTach trace for me, you can use Imageshak or the like if you don't have any personal hosting.
 
Hi! Thanks for the reply. Great memory! Here are a bunch of screen shots:

My current RAID5 performance:

RAID5.png


The performance I got when I had my 4 disks in a RAID0 array:

RAID0.png


The performance I currently get from a Seagate 7200.10 750GB drive:

750GB.png


continued...
 
The settings look OK, NCQ off is correct.

The dip at the start of the RAID0 trace is a bit odd though as are the burst rates, they seem a bit low. I'm wondering if one of the drives isn't performing as well as the rest, unfortunately that'll be a pain in the rear to test.

You've got the caching at the array level set to Write-Back rather than Write-Through?

Oh, the Seagate's fine btw, you might want to whip the jumper off the next time your inside the PC to let it go at SATA2 rates but that'll only affect the burst rate, you won't notice any real world difference.
 
Thanks loads for the quick reply.

rpstewart said:
I'm wondering if one of the drives isn't performing as well as the rest, unfortunately that'll be a pain in the rear to test.
Click to expand...

Is there any way to test individual drives without having to destroy the RAID5 array?

rpstewart said:
You've got the caching at the array level set to Write-Back rather than Write-Through?
Click to expand...

Yes, caching is set to write-back.

edit I've done a few more tests and experiments. Firstly, I turned off all overclocking settings in my mobo's BIOS (i.e. my machine is now running at stock speeds). That didn't improve anything (if anything I've reduced my read rates by a few MB/sec).

I've also checked the SMART status for each drive in the RAID5 array and each drive checks out OK.

edit2 I've now looked on the WD website for firmware and software tools. There's no firmware available for my WD5000AAKS drives (although there is firmware available for the WDxxxxYS drives). I've tried running the WD tools application but that couldn't see the WD drives attached to my RAID controller.

edit3 I checked on highpoint's site and just in the last few days they have released BIOS version 1.12. I installed the new BIOS and re-tested my RAID array but it hasn't increased the spead at all. Also, I'm running the most up to date Mobo BIOS and the most up-to-date nForce drivers.

Thanks,
Jack
 
Last edited:
You'll be able to test the drives individually with the array running with no redundancy while you do it - disconnect one disk from the RAID controller and attach it to a mobo SATA port then run HDTach, it shouldn't need formatted. You might be lucky and be able to put it back into the array without rebuilding, worst case however is it takes 8hrs or so to rebuild before you can do the next one.
 
Thanks for the reply. Good point about taking out one drive at a time and testing it as a "normal" drive.

I'll make sure my backup is completely up to date (which will take about a day) and then I'll test each drive. Oh, what fun computers are ;-|

Thanks,
Jack
 
Hi again,

So - I've now backed up all my data and I've tested each of my 4 500GB drives individually. Here are the results:

first_2.png


second_2.png


To my eye, all drives look like they're performing about equally.

I've run out of ideas as to why my RAID5 array is running so slowly. Any clues?!?

Many thanks,
Jack
 
The results above were done using the "short test" on HDtach.

The results below were done using the "long test":

first_2_longtest.png


second_2_longtest.png


The drives are connected to my mobo's SATA connector.

I guess my next experiment should be to re-connect the drives to the Highpoint RAID adapter, destroy the RAID5 array, re-configure each drive as a stand-alone drive and then re-test.

Thanks,
Jack
 
dan_aka_jack said:
I guess my next experiment should be to re-connect the drives to the Highpoint RAID adapter, destroy the RAID5 array, re-configure each drive as a stand-alone drive and then re-test.
Click to expand...
I'm starting to run out of ideas too. The above might be a worthwhile exercise but it's a real pain in the rear to have to nuke the array...
 
rpstewart said:
I'm starting to run out of ideas too. The above might be a worthwhile exercise but it's a real pain in the rear to have to nuke the array...
Click to expand...

True. But there's definitely something fishy going on. My RAID5 read figures are LOWER than I get for a single drive.

And now's a good time to destroy the array because I have everything backed up.

Oh, you were right about not having to re-build the RAID5 array after testing each drive. I just put the RAID5 array back together and the array started without any complaints at all ;-) Shame I have to kill it now...

Please let me know if you can think of any more tests...

Jack
 
Quick update:

I destroyed my RAID5 array and then I configured each disk as a single-disk JBOD. Each disk performed identically compared to when they were connected to the mobo's SATA connectors.

Then I tried building a RAID0 array using my highpoint software. I tried a 4-disk array and couldn't get it to perform above 160MB/sec (it should get to more like 250MB/sec). I tried building two 2-disk RAID0 arrays and they both performed identically. I'm starting to believe that maybe the RAID performance of my cheapo HighPoint RocketRAID 2300 is just a bit crap.

Then I tried making a WinXP software RAID0 array with my 4 disks. Bam! 250MB/sec read speeds with no tinkering. Right now I'm formatting a software RAID5 array and we'll see what performance that delivers...

http://tomshardware.co.uk/2004/11/19/using_windowsxp_to_make_raid_5_happen

Does anyone have any experience with WinXP software RAID setups? The TomsHardware guide above recommends the setup.

Jack
 
Last edited:
Hmm. It's a little hard to benchmark my new software RAID5 array because neither HDtach nor HDtune recognise the WinXP RAID array. And I don't understand IOmeter. And XP's "Performance" application spits out some damn wierd numbers sometimes.

Anyway... I think the bottom line is that my software RAID5 setup is s l o w. I'm getting write speeds of about 10MBytes/sec and read speads of about 60MBytes/sec.

Back to my highpoint RAID5 array!

Jack
 
BillytheImpaler said:
Might part of the problem be CPU overhead as the CPU has to do the calculations to for the cheapo RAID card?
Click to expand...
Not for reads which is where the problem is, they come straight off the disks RAID0 style. Writes however, yes, CPU power will limit the speeds.
 
I've settled for a slightly bizzar config which is a RAID10 array where the highpoint card makes two RAID1 (mirrored) arrays which are then stripped together by WinXP. This array averages about 120MB/sec reads. I think this is the highest performace RAID array I can build which includes redundancy. (The highpoint card can do RAID10 on its own but that can only get a bit of 100MB/sec).

If I had Linux installed on my PC then I'd love to see what a Linux software RAID setup could do... but I don't have Linux installed and I can't really justify spending that long tinkering.
 
You must log in or register to reply here.
Back
Top Bottom