You can move the pagefile to the drive that you like without any problem.

Microsoft's own recommendations about the subject for XP can be found here:

   http://support.microsoft.com/kb/314482

As XP and 2003 are roughly the same generation, I'd assume the same applies to 2003. This is also suggested on many other pages.

So according to the article above, you need to have a small pagefile on the boot partition (I assume that's C:\ in your case, as usual) BUT ONLY IF you want to be able to save the memory dumps and debugging information in the case of a system crash. If you are not worried or interested about that, then you can move the pagefile completely to another drive.

I've moved page files perfectly fine without one on the main C:\ drive.  You should be safe to do so.

I believe that that pagefile on the C:\ drive needs to be at least equal to or greater than the amount of physical memory you have.  Otherwise, you will not be able to dump the entire contents of the memory during a BSOD.  You will not get a good memory dump without a pagefile on the C:\ drive!

technicalgadget: That's probably true but the question is: are the complete memory dumps or even the alternative minidumps really interesting or useful? In most cases if the system runs stable 99.9 % of the time I'd say they are not.

100% YES

Consider;
Patch Management
Service Pack Upgrades
Disk or File Corruption due to unstable/spking power supply
Data Loss due to No Battery Backed Write/Read Cache
Failing Hardware such as memory or even CPU
Upgrade of Hardware Drivers

You might never know the true cause of a BSOD without a good memory dump.  I'd say in many cases, an administrator resorts to rebuilding the OS or reimaging a server from backup before getting to root cause analysis and even with a good memory dump, you still might never get a good root cause analysis, but something is better than nothing.

BOSS:  What happened?
ADMIN: I don't know
BOSS: Why?
ADMIN: I stayed overnight and just rebuilt the server from scratch
BOSS: <shakes head>

If you run a data center, then getting to root cause analysis is even more critical as the same issue could affect multiple servers.

Or...

BOSS:  What happened?
ADMIN: I don't know
BOSS: Why?
ADMIN: I've used the past two weeks trying to reproduce the problem, studying these memory dumps and googling for answers about this OS that just happens to blow apart once a year. Of course, I could've just rebuilt it in one working day like most people do...
BOSS: <shakes head>

:)

If you want a FULL memory dump then yes, you need a pagefile the size of RAM + a few MB (don't remember off the top of my head) on the C Drive.

That said, I very much disagree with Technicalgadget's opinion.  In 17 years as an IT professional, I've NEVER had to analyze a full memory dump, this is over hundreds of server installs and thousands of workstation installs.  Could it happen?  SURE.  Is it likely?  NOT AT ALL.  small memory dumps USUALLY provide enough information to know the cause of the problem.  And the idea that an administrator is going to spend HOURS analyzing a dump is just not practical.  There COULD be instances, here and there, but IN GENERAL, in a large environment, redundancies and imaging speed server deployment and where downtime costs money, determining root cause, while a noble idea, is not generally worth it, economically speaking.  (I used to love to fight the problem until I determined what it was... but over time, I've learned how much my time and my client's time is worth... doing an analysis of a problem is often not worth it.

I DO think it's important to point out the issues with moving it off the C: drive, BUT, I do believe that moving it off the C: drive, in MOST cases, is a very appropriate action.

@arimc
LOL, I hear you.  I  do what most do when a BSOD occurs, call up Premier Support and have them analyze it.  No need to spend countless hours finding symbol libraries, connecting debuggers, trying to reproduce the error, etc.

@leew
I respectfully disagree.  Having options available to me to get to root cause analysis is certainly better than having no options at all.  I want to have the choice to get to root cause analysis, not restore from backup every time there is a problem.  Your philosophy is financially motivated and while I get that, learning should be a never ending process, not a series of SOP's.

> Your philosophy is financially motivated and while I get that,
> learning should be a never ending process, not a series of SOP's.

I agree... but SURVIVAL is also a never ending process... well... until you die.  I cannot bill my client for 6 hours of analysis of a crash when I can re-image their computer in 30 minutes.  The client is unable to use the system while it is down, so it's not just paying me - it's LOST PRODUCTIVITY for them.  And even if I offer to bill them JUST for the 30 minutes it would otherwise take, that's 5.5 hours of my time I can't be billing another client or spending with my family.

Frankly, beyond basic analysis of crash dumps, I've never done a full analysis of a crash dump. I've never had MS ask for a full dump.  If you've got the space, go ahead - in this instance, he clearly doesn't have the space and I don't think risking corruption with repartitioning is a wise idea.

I love EE .  Your feedback far exceeded my expectations.  I'll have to say, in the 12 years I've managed probably 100-150 server installations and only once leveraged a memory dump for troubleshooting.  However,  I see their value in certain situations and it is 'best'  to have one as an option in troubleshooting.    Would it be appropriate to leave a small 600MB pagefile for 'minidumps' on C:\ and create a much larger one (such as 6GB) on another partition with sufficient free space?  The server has 4GB RAM.  Thanks -

The minidumps aren't going to be that big - that's why I say to leave a minimal 128 MB pagefile.

Highly recommend the article - a very informative read and helped in my situation.  thanks for the effort by all who commented -