Virtual Memory Settings Question

[muthabored]

Hi, Guys:

I've just installed 1 GIG of DDR RAM in one of my PCs. The PC is running Windows 2000 Pro, SP3 and my HDD is 40GB. I've never tinkered around with the Virt. Memory settings I know where to find the settings (System Prop/Advanced/Performance Opts/Change) but I was told that I should set the page file size to the amount of physical RAM that I have. It shows my initial size to be 192MB. It shows the maximum size to be 384MB. Are these sizes from the HDD or the memory? Also, which one (initial size or max size) needs to be changed to 1 Gig? TIA!

 

[euston]

Microsoft's rule of thumb is to set VM size at system installation time to an initial size of 1.5 x RAM with a maximum size of 3 x RAM.

I like to put the paging file on a second HDD (I always use 2 HDDs) as this improves performance cos 2 separate sets of read-write heads are working independently of each other. This reduces head seek conflicts.

 

[joecastaldo]

Microsoft's rule of thumb is to set VM size at system installation time to an initial size of 1.5 x RAM. However working closely with Microsoft over the past few years set the max and min to the same size. This will eliminate the resources changing the size of the page file. If this is a Database or Exchange server it should be 2 * RAM.

 

[muthabored]

Thank you for the information/replies

 

[devilman0]

Less is more, ignore m$, reduce page file to almost nill. Open up all the programs you use, get them doing something, then see how much mem you have, this will be your starting memory, subtract your ram from this, this is your page file size, add extra 10% for good measure. I followed M$'s recomendations, well when i saw another user doing what i just said i tried it and my computer gained a 25% performace increase. Windows tries to put everything it can to the page file, it may not even use the system ram until the page file is full (is actually documented somewhere, but i can't find the link, so just use as "hear say".)


Thanks,
James

"Make it idiot-proof and someone will make a better idiot." ~bumper sticker

 

[cdogg]

James,
That theory is not entirely true. The page file will be used in conjunction with RAM, not "before" it. Think of it as an extra container to store less active items and save space in RAM for more important operations.

I would also question your claim of 25% performance increase by using a small page file. Maybe that was the case in your situation, but what benchmarks (if any) did you run to show that? On workstations, you may or may not notice difference by having a small page file. On servers (database, application, or domain), it would be a disaster. Windows needs to be able to effectively dump memory into the page file to correct errors and also in the event of a crash. Having a small page file would limit this ability and ultimately slow performance during heavy access times.

Again, if this is just a workstation, then I would try it both ways and see. I doubt it would matter much either way...




~cdogg

"The secret to creativity is knowing how to hide your sources."
- A. Einstein

 

[devilman0]

True windows does put everything to disk first, but when it can, it will. the benchmark i used was a 3dstudio max v4 with a 2min animation (2min * 60sec * 30fps = 10800frames) with raytracing and reflection(water) and smoke effects. Ran this with a page file of 1gb, recorded the results (did this over a year ago, but) then ran it with a page file of about 128 or around there abouts and was 25% to 50% faster. (ran tests 20 times to get a "real" result) Seti@home had an decrease in time that it took to "crunch" a WU. These instance shows that windows tries to put a lot to disk, first or last it really doesn't matter, in the end the more windows uses the slow ide or scsi channel, the longer it takes, if you have a heavely bogged down system with just writing info about databases and such then your system will slow down more and more. In my case i have an ide 133 40gb 10krpm. Putting the page file on an "unused" disk is a waste of money, i would put that money into some good memory (hey, memory is cheeper than "unused" hard disks) put those hard disks to work storing stuff, use memory for swaping info, it is much faster, more efficent, and is cheeper in the long run. Everyone who knows high performace knows caching to disk is slow, in any case, more system memory is better. True if you have a server app that loads large amounts of memory you would need a fairly large page file. I run iis, an email server, AD, FTP, antivirus, tape backup, etc and only have a page file of 256 with memory @ 256. It all depends on your tastes, you like going slow, make that page file bigger and leave the settings @ default, you like going fast, get your hands dirty under the hood.

all this is just my 2 cents.



Thanks,
James

"Make it idiot-proof and someone will make a better idiot." ~bumper sticker

 

[cdogg]

No offense James, but I don't think you've ever been in a corporate environment running PDC's, BDC's, Exchange Servers, SMS, proxies, etc.

If we were to shrink the swapfile to a size smaller than the amount of RAM on each server, the result would be a slower, less reliable environment - not a faster one. Your example is different because it's isolated free of any dependencies. It would be naive, to say the least, for you to assume that your suggestion would make everything quicker. Everyone seems to believe that by making a change, they are "tweaking their system" to run faster. "Getting your hands dirty under the hood" can often backfire.

Using your reasoning about using less cache (which every "high performance" expert knows), why have any cache? Why not set it to 0?

The answer is obvious. Although the reasoning in your mind says "less cache is better", Windows central kernel operations rely on it. Instead of telling people to ALWAYS use less cache, you should be educating them to decide the right amount for themselves.

Bottom line: Different environments call for different settings.




~cdogg

"The secret to creativity is knowing how to hide your sources."
- A. Einstein