Windows server 2003 raid 5 becoming very slow

[lengis]

Having 3 1Tb disks on my server in a raid 5. Recently this volume has become very slow, both read and write are about 4 times slower than it used to be. Locally copying a 1 Gb file to this volume was like 30-60 sec before and now it's like 3-4 mins... About 50% filled Any ideas???

Cheers

/Lengis

 

[sub mesa]

RAID5 is a theory, you did not specify what kind of RAID implementation you use. For example a hardware RAID add-on card, or using Windows' internal RAID engine using dynamic disk configuration. Windows' support of RAID5 is very poor and it should not be used by those who require either performance or reliability.

Instead, software RAID-engines under BSD and Linux are very advanced, and can yield much higher RAID5 performance than Windows' own implementation can ever provide.

A reason for a RAID5 to become alot slower is usually because the array has become degraded, due to one of the disks malfunctioning. You can check if this is the case, with any management utility your RAID engine has.

 

[lengis]

Sorry, should of course mentioned that it's a win 2003 srv internal raid. None of the disks are faulty as far as i can see. Nothing shown in the disk management tool or smart counters. I'm well aware that the internal windows raid isn't state of the art but it was sufficient for my needs. Strange though with this sudden decrease in performance...

/Lengis

 

[sminlal]

Can you post a screenshot of what you see in Disk Manager? What you're describing sounds like a failed volume.

 

[lengis]

Free Image Hosting by ImageBam.com

/Lengis

 

[sminlal]

Yep, that looks pretty nominal all right. I'm wondering if there's something else accessing the volume at the same time that's causing contention. Have you tried using PerfMon to see what the I/O queue length is for the disk or if there are other I/O intensive processes running?

 

[rand_79]

try running a benchmark on it also and screenshot the results.

something like hd-tune or hdd-tach. or something similar

 

[lengis]

Having checkdisk running atm, not at the fastest pace i might add

Will try running a benchmark as soon as it's done...

/Lengis

 

[lengis]

Guess it's a faulty drive after all, even though windows disk management tool or the s.m.a.r.t counters don't say so....



Free Image Hosting by ImageBam.com

/Lengis

 

[sub mesa]

I can't view that image, its sent by the webserver with a MIME-type of "image/$file_ext" which is a PHP programming error.

Could you pick a different host?

 

[lengis]

http://www.imagebam.com/image/acf2e550512131 should be working

 

[sub mesa]

I'm sorry but you did not complete the benchmark, which is why some data is missing. Its also hard to read because its a cropped picture (made smaller than the original).

If you are able to, rerun the benchmark, especially the disk that had 3.3MB/s speed, it might be using PIO transfer instead of DMA transfer. If this is the case, you should see a very high CPU usage indication, for a dualcore its close to 50% for a quadcore close to 25%. By letting the benchmark finish it will show this data.

Please also crop the picture by a good program so your website doesn't resize itself, because it uses bad quality algoritms for this.

 

[lengis]

Sorry about the poor and incomplete pics, i was so amazed by the bad results that just posted it without letting the benchmark finish...

Excuse my limited knowledge about this stuff. These are all sata disks attached to the same controller, i know how to change transfer mode on an ide controller, can you do this on a sata?

Here are the complete results:

http://img169.imageshack.us/gal.php?g=disk1.png

Hope the image quality are better )

I will also open the lid to check all cables asap.

Sub mesa, thanks for taking the time looking in to this.

/Lengis

 

[rand_79]

are those 3 pictures the 3 different drives?

if so then yes the first one has issues.

 

[sub mesa]

This disk has indeed issues. I was correct about my earlier presumption, that one of your disks had a PIO issue. This means the transfer of data won't go via direct memory access or DMA, but the CPU gets to do all the work and it will completely saturate one CPU core when reading or writing.

As in your case i see a value close to 25%, suspecting me you have a quad core CPU. Try gathering some SMART data from this drive, and check its surface, replace the data cable used and if that doesn't help, you can try this: go to device manager, look for the sata controller that holds the disk (dunno how windows names them) and remove/uninstall/delete it. Pressing the delete key when selecting the controller will do. By removing the controller (not the disk!) from device manager it will 'reset' next time when you reboot. If you rerun the test on that flaky drive and results return to normal; your issue is fixed again!

Good luck

 

[lengis]

Sorted

Did as Sub mesa suggested and deleted the sata controller from device manager and rebooted twice to have it "installed" again. Running the benchmark test again after that showed same permformance on all 3 drives ~100 MB/s... Yippe

I was about to have the drive exhanged by warranty.

I guess i have to live with not knowing why the controller was messed up, that's kind of default when dealing with microsoft...

Sub Mesa, your my idol. Thanks for helping me out on this one.

Regards

Lengis

 

[billu10]

I also have the same problem. I do not have my system information, but it is an AMD Phenom II X4 805 quad
core processor with 4GB RAM and 2x 1.5TB + 1x 1TB Seagate HDDs configured as software RAID 5 on Windows Server 2003. During writes, it transfers data over a network at good speeds until about 2GB and then slows down miserably. My CPU is hardly utilized.

As suggested by Sub mesa, I cannot find the SATA controller in the Deice Manager. My north-bridge and south-bridge are AMD A760G and SB710 respectively. The SATA controller is on the south-bridge AMD SB710. But I am not using its RAID since it does not have RAID 5.

Any thoughts on why the transfer speeds reduce so much after about 2GB data writing? Also, how can I disable the SATA controller ports and have them reset upon reboot as suggested by Sub mesa, as I cannot see them in the device manager?

Any input will be highly appreciated.