Topic awaiting preservation: swapspace partition (Page 1 of 1)

So here and there, mostly everyone recommends you to make a partition for your windows pagefile.

Now, about the size of that page file, and if it is to be set to system managed or to a manual fixed size, opinions diverge. Most go for the manual option and set both maximum and minimum to the system recommended page file size.

But about the partition itself, there is little information. A program i used, made an automatic partition for the page file, and made it a fat32 with 4k cluster size.

I was thinking if this was really a good choice, being the page file such a large file, so i changed it to ntfs, with a cluster size of 32k and noticed a bit of a performance increase.

Has anyone got some useful information about this, or can direct me to some?

edit: oh, and about the parition, should it be logical or primary, or that aspect doesnt matter?

Thanks in advanced,
arthemis

(Edited by Arthemis on 11-27-2005 03:28)

very good question. I don't know that much about the swap partition so after reading this post, I was intrigued so I found this.

It seemed to be pretty good but I will have to wait to see how much I really understood of what he was talking about LOL.

Later,

C:\

thanks cprompt. Interesting page.

I read it and got a review on my current knowledge, with a few new pointers. Still, it all refers more or less to the nature of the pagefile itself.
Except for the fact that this guide points out that it should be ordered first where possible, little is said about the swapspace partition.

i think that for my question only the

quote:

---

"A STATIC swap file on a separate partition, but on the same physical hard drive as Windows. "

---

configuration is of interest.
*smile*

(Edited by Arthemis on 11-27-2005 06:42)

A while back I setup photoshop to keep it's own swapfile on a separate patrition on a separate harddrive. That made a lot of difference compared with having it on the same drive.

Havn't played around with the windows swap though, but it seems reasonable that if the swap can be accessed on a different HD while winblows does the rest of its thing on another HD it should be faster...

As for a separate partition on the same HD... Havn't got anything to back this up, but no matter how fast a HD is, it still can't access more than one place at the same time which would speak for the 2-drive alternative.
/Dan

{cell 260} {Blog}
-{"Theories without facts are just religions...?}-

Moving the paging file to another partition doesn't do as much as people think. It's still on the same drive, so the same read/write heads are doing the work as they would if the paging file was on the same partition as the OS files.

You'd see far more benefits by moving to another physical drive.

quote:

---

Moving the paging file to another partition doesn't do as much as people think. It's still on the same drive, so the same read/write heads are doing the work as they would if the paging file was on the same partition as the OS files.

You'd see far more benefits by moving to another physical drive.

---

True. I think that mostly we all agree is that the best solution is to have the swap file distributed among the first partitions of 2 (or more) non-os fast access hard drives.

But. What should those partitions be like? I'm only portraying the simple configuration of a swap file in another partition, in the same hard disk as windows, to simplify matters. The question sustains itself, whichever configuration you chose, and it is on those configurations of less performance, that it is most likely of most importance.

i was researching this a while back but more for win98 'cause i'm still stuck mostly on an old old box

but i found this reference in my notes for XP

http://aumha.org/win5/a/xpvm.htm

my last post should read like this

quote:

---

True. I think that mostly we all agree that the best solution is to have the swap file distributed among the first partitions of 2 (or more) non-os fast access hard drives.

But. What should those partitions be like? I'm only portraying the simple configuration of a swap file in another partition, in the same hard disk as windows, to simplify matters. The question sustains itself whichever configuration you chose, and it is on those configurations of less performance, that it is most likely of most importance.

---

outcydr, the page your provided contained this

quote:

---

Should the drive have a big cluster size?

While there are reports that in Windows 95 higher performance can be obtained by having the swap file on a drive with 32K clusters, in Windows XP the best performance is obtained with 4K ones ? the normal size in NTFS and in FAT 32 partitions smaller than 8GB. This then matches the size of the page the processor uses in RAM to the size of the clusters, so that transfers may be made direct from file to RAM without any need for intermediate buffering.

---

i also found the same 4k ntfs recommendadion, in other pages.

Now, even if this time, a little explanation for the cluster size choice was given, there is no assurance of this being really the best choice, or the current trend.

let's see... a cluster is the minimum data that can be addressed on a file system - you could read half a cluster, but you can't read the second half without receiving (and possibly discarding) the first half as well.

Windows on x86 systems uses 4kb page size - the minimum memory amount that can be swapped out.
So, a cluster size of 4kb is likely to be ideal.
Fat or NTFS probably does not matter much, since there is only one file - and buffering probably isn't a concern.

In the end it boils down to:
Get a benchmark program and try it out.
It might very well depend on your physical disk layout - some disks have 'faster zones' and a custom swap partition might let you control where that's located. That's the only advantage against a fixed minimum size swap file on any old partition I can see.
Oh, and before you do all this, check your system performance monitor for swap activity... if your system ain't swapping, you're never going to notice a difference ;-)

so long,

->Tyberius Prime

thanx, that was helpful

~this is not a signature~