I have a customer whom was supplied an SBS box by thier previous supplier. It is only a 5 user system but the system runs incredibly slow, particulary Exchange.
Although the specification of the box is reasonable(Pentium 4 2.8GHz and 1.5GB Ram) performance has always been bad since the day it was installed. I have found other users with the same problem which is related to a particular IDE raid controller that HP/Compaq included on thier cheaper range of servers.
Also the server has extremly high CPU usage whenever the disks are being accessed. The beleif is that this CPU usage is in someway down to the hardware/software combined raid this system provides.
Having done some benchmarks from within Windows on the disk subsystem it is obviously the problem, thier server scores significantly less well than my laptop which has a much slower hard disk.
Because they have only 2 disks which make up a mirror set, I thought of this idea to get a speed improvment on the cheap hopefully.
Now I dont know about anyone else but this sounds like a plan which is sure to go wrong to me, but it maybe seems slighty easier than doing a bare metal restore......
Thoughts anyone?
Paulie,
1st off, always make a full backup - better to spend a little while doing this, rather than spending your time recovering .
Sometimes when I have seen this issue, it actually relates to the fact that disk caching is disabled on a domain controller if the disk drivers do not have special power backups to ensure the integrity of the data - go have a look in device manager to check 1st.
2nd, if it is not turned on, but you can turn it on and performance improves, it may be reset at reboot, in which a little registry patch is needed. I will see if I can find it. I would explore this before I went any further - and you might find that moving to the onboard IDE does not solve this problem - possibly something to test before you do too much work (perhaps disable 1 disk, move the other to the onboard IDE and see what performance does)
2nd, this is not a guaranteed statement, so I don't want to be held responsible if it all goes wrong
I did this a little while back, moving from hardware RAID to Windows RAID, by moving disks and joining then to the set. Asssumung the disks are JUST mirrored and no striping is involved, and that there is no special headers etc on them, you can do what you want and the process of moving one disk then the other should work. In fact, putting both disks straight onto the IDE controller might even work as Windows will not longer see them as RAID'd and will boot of the 1st drive.
It is important thta you format the 2nd disk to stop any signiature issues.
There are two differences in the system that you need to be aware of once it is done. 1) performance will be impacted by using software RAID, not much at all, but it would show on a numerical benchmark (I have never noticed it using a system with this config, but be warned). 2) should the 1st disk fail, while all data will be fine, the machine will need to be shutdown to replace the disk and then rebooted. However if the machine shutsdown in this state and the disk is not replaced, the bios may well see the bad disk as the boot disk and fail to boot up until the bios is told to boot of the 2nd disk.
Does this make sense?
ttfn
David
David,
This makes good sense and it is helpful to know you have done it before! Couple of things:
I cannot change the write caching status of the disks from within Windows but I can change the setting in the BIOS of the raid controller. It is somewhat faster with it on than without it, but still is very slow.
As you say the CPU overhead of software raid never seems to be noticable, I have used it quite a few times and always found it to work well.
Here is the link to the other systems experiencing the same problem:
http://forums2.itrc.hp.com/service/forums/questionanswer.do?threadId=491669
Some of them seem to have come to a resolution but having done all the bios and driver upgrades possible it seems that moving the disks to a different controller is the only real way forward.
I am probably going to do this on Wednesday night so I will let you know how it goes! Thank you for the help and advice :)
On the disk caching side - this is unusual, I am used to there being a tab on the disk in device manager (Policies) that enables write caching & additional perf features. At the very least, this will probably go away on reboot when moving to IDE based controller.
The thing here is to have a go and see what works, but don't get too upset if things go wrong.
On the disk perf issue without windows write caching, let me know if this turns out to be an issue and I will try to find the details to fix it. (i.e. if perf is still bad without the settings ticked when an IDE drive, but once ticked, perf works again)
ttfn - off to be part of the family
I was perhaps a little misleading in my response to you before. The policies tab is there, but the options are disabled, see screenshot here:
www.accendo.co.uk/disk.JPG
Maybe because windows does not have that leve of control of the disks because they are part of a hardware array?
Heya there Paulie,
That tab is disabled as the controller is a hardware one, and the write behind cache settings are set from the hardware BIOS.
It does improve IDE performance, but for various Gods sakes, don't do it without a UPS on the server without software to tell the server to shut down.
I've had a few play servers kill themselves down to this turned on and no UPS there to support the cache.
(Or buy a RAID controller with a battery on it!)
Horrah!
Moved to the standard IDE controller tonight and used Windows mirroring. All went without a hitch and performance has increased by a HUGE amount.
Thanks for the advice chaps.
Paul
Paul,
Just rememebr what Jules said - the chances of disk corruption in the event of power failure etc is much higher now, so those backups need to be even more solid until you can change that.
thanks
The system is already on a UPS so not too much of a worry. The chances of disk corruption are no higher now than there were before. Write caching was enabled beforehand, but the performance was bad.
Though obviously the PSU in the server could go pop and a UPS wouldn't make any difference at all.
(c)David Overton 2006-23