Swap File Optimizations?

fastswap asks: "I've got a pretty standard computer with reasonably fast drives. I've got an old 2GB-but-fast drive, and a spare channel on the motherboard. Does it make sense to install the 2GB drive on its own controller and use it for a dedicated, fixed swap file? I figure if the computer's using the swap file, then in the current setup with the swap file on the primary controller, then it's contributing to hard drive thrash exactly when one doesn't want it to (i.e. when the machine needs the swap file). If it is better to have a dedicated swap file on its own controller, is the same true for other operating systems with similar approaches to virtual memory? Since drive space is so cheap now, should the swap file be fixed size anyway rather than letting Windows suddenly get the urge to resize the thing?"

Dedicated

[yosemite]

It is better

Re:swapping?

[Hythlodaeus]

As is often the case, games are making some of the greatest demands on hardware. Many current games fill 300 or more MB by themselves, and right now I'm sitting on 225 MB with Mozilla, Word, and mIRC open.

Separate swap under linux.

[derrith]

I've done this under linux, say get a 2-6GB drive and use it as dedicated swap. I tend to do this with scsi servers when I'm patching together old gear. Say an 18GB root drive and then the smaller drive as dedicated swap. leaves everything open. And if the swap drive does get thrashed, no big deal. It's quite effective and works well in my experience.

Dedicated is better; linux lets you RAID it

[Anonymous Coward]

Dedicated is better.

However, you will never get true swap performance using Windows.

To do that you need a real operating system. Linux will let you put one swap partition on each controller, set them to the same priority, and it will automatically spread the access between them, getting a RAID-like speedup in your swap access times.

Also, remember to put swap partitions (if you are using files you are hopelessly fucked) on the end of the disk, so that they will be on the outer sectors where the transfer rate is fastest.

Linux Swap Space Mini-HOWTO

[sICE]

This [ibiblio.org] Linux Mini-HOWTO might be of interrested to some /.ers, it describes how to share your Linux swap partition with Windows.

Absolutely!!!!

[OC_Wanderer]

From Windows XP back to Windows 95, gurus have suggested a fixed size of 2 or 2.5 times the size of your RAM. I keep it at 3 times, because I have CRS disease and can't remember the exact size. Better safe than sorry, since I have the room.

Swapping on a separate drive is faster than swapping on the same drive. I've tested that. I also put the "temp" directories on the separate drive, as well as the data directories for my applications. This includeds the mailbox for Outlook Express and the temporary internet files for Internet Explorer.

There's a big bonus to setting up like this, besides performance. There's less to backup from C: drive!

[Contrary to popular belief, not all nerds and geeks use OSS.]

two drives and 1 controller, with striped swap.

[Elivs]

I have two ide drives and 512M ram. I have the first 512M on each drive as a swap with linux stripping the swap. (See "man 2 swapon", "man swapon", "man 5 fstab" and set the priorities of each partion to be the same) I did this on the assupmtion that the bottleneck is likely to be disks's read/write speed, not the controller.

Like you I'm also not sure if it makes much difference but my system certainly seems to often be swap limited. I currently have KDE3, several gnome apps, a browsers with 4 windows (20+tabs), 2 virtual desktops, and I often use octave to process high resolution images. Changing from one app to another can cause the machine to swap for a few seconds if I've haven't used the first app in a few hours/days.

Elivs

Clearly if I used windows I wouldn't have these problems as I could never leave apps idle for days while doing another task.

/me Ducks as an "MS wireless mouse" flies towards me...

If security is a concern

[Anonymous Coward]

If security is a concern, set Windows to use a fixed swap size, i.e. the minumum and maximum values should be identical. Then, install Eraser [heidi.ie] (GPL), which offers the ability to wipe the swap file during the reboot process.

Re:Fixed size...

[zatz]

I think the recommendation of "at least twice physical memory" makes sense only with a VM system which employs a one-to-one mapping between allocated pages and disk blocks. I do not believe either Linux or Windows works this way.

Re:Fixed size...

[Brandybuck]

I've never tried putting the swap on its own channel or controller - or even on its own drive - under any OS. Like you, I'd be interested in hearing whether or not this is worth the trouble.

I still do this, but with 1G of RAM, I never swap anymore. Back back when I had a 100MHz system and 32M RAM, putting the swap on another harddrive made a significant difference. That was with Linux. Since Windows uses a swap file instead of a raw partition, so it might not make much of a difference.

Re:Fixed size...

[WSSA]

Don't let Windows resize the swapfile - that's a surefire path to fragmentation, fragmentation = slowness.

One Word Answer: No

[silverfuck]

Okay, so this is too late for all but the most sad of slashdotters to read it, but here goes:

If the drive is 2GB, then don't be so sure that it is fast - it may have been when it was bought, but that was 6 or so years ago at least. I would be very suprised indeed to see more than 4-5MB/s sustained read and 2-3 write; there have been a lot of advances in the last few years.

My current setup (1GB physical RAM) has 2GB set aside for each of Win2k and Linux in seperate partitions right in the middle (this will speed up average access times as the heads will have the least far to travel on average from any random point over the platters) of the raid array (and hench middle of both disks, as it is RAID-0), which I know to be fast - benchmarking has pegged it at greater than 110MB/s sustained. Windows will hit the swapfile no matter what (just try setting the swap to 0, even on a well-heeled system, and watch it complain at bootup/logon), so it gets 512MB to play with just at bootup and can go all the way to the end of it's swap partition if it wants. Linux, well, that's another story (currently support for the raid array is patchy, so not running linux - the partitions are still there, though, waiting for filsystems!), but as everybody knows, linux is very aggressive about swapping stuff out and using physical RAM as a disk cache, so again I expect it to hit the swapfile after a few days (hours?) running, but be perfectly happy with 2GB.

Re:Linux Swap Space Mini-HOWTO

[Anonymous Coward]

Caution advised if you want to do this and use Software Suspend [slashdot.org] on Linux. (i.e make sure it's not your swsusp swap space that's going to get overwritten if you suspend Linux and boot into Windows).

Performance considerations

[jcasey]

If thrashing is causing degradation, I would seriously consider increasing RAM before improving on the swap drive.

Check the performance specs for that 2gig drive first. If you are connecting an older, slower drive, you may actually worsen performance. For best performance, use a drive that can supports whatever performance features your mobo offers ( UDMA-66, Serial ATA, etc... )

IF using Windows 2000/XP you can spread your page file accross multiple hard drives.

Putting swap in a RAM disk makes no sense

[mlq]

Putting swap in a RAM disk makes no sense what-so-ever.

Re:Yes, seperate drive and fixed size

[Quarters]

It is important to note that WindowsXP will use the page file whether you've got plenty of RAM or not.

You can instruct XP (and probably 2K) to not page the executive and to use more memory as cache space. This reduces the amount of paging significantly.

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Contro l\Session Manager\Memory Management
*Change DisablePagingExecutive to 1
*Change LargeSystemCache to 1
*Reboot

The obvious solution

[Anonymous Coward]

Putting more ram in the machine instead of adding another power consuming hard drive will give you a much better performance boost.

Re:Yes, seperate drive and fixed size

[Zeriel]

Well, back in the good old days when 4MB was "more RAM", you generally only had a machine with relatively large amounts of RAM if you were processing a similarly large amount of data.

Joe User would have a machine with 4MB RAM and 8MB swap for his word processing and Ultima 2 or whatever.

Stan Scientific would have a machine with 32MB RAM and 64MB swap because he probably was going to eventually have to deal with datasets larger than 32MB (if you've done ANY scientific computing over historical datasets, you know what I mean).

Basically, more RAM implies you should be swapping less, for a home system.
But for a server or high-end processing computer, more RAM implies you need vast amounts of RAM in general, and swap doesn't hurt.

Re:Putting swap in a RAM disk makes no sense

[Strange Ranger]

Parent is not informative. Provocative maybe. In order to be informative, the parent's statement needs a supporting explanation, including what OS.

RAM disks are fast, Windows requires swap no matter your physical RAM size, so why not put it on a RAM disk?

What are us dummies missing mlq? Please elaborate.

Re:Yes, seperate drive and fixed size

[JeFurry]

You can instruct XP (and probably 2K) to not page the executive and to use more memory as cache space. This reduces the amount of paging significantly.
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Contro l\Session Manager\Memory Management
*Change DisablePagingExecutive to 1
*Change LargeSystemCache to 1
*Reboot

True, but doing this disables standby and hibernate modes, since the kernel can't be unloaded any more. If that's not a problem for you, go ahead of course, but it's worth being aware. I did this, and kept finding my system going into standby on request, but never resuming, and it took me ages to find out why...

Re:Fixed size...

[Just Some Guy]

FreeBSD, on the other hand, explicitly recommends having at least twice as much swap as real memory. From tuning(7):

The kernel's VM paging algorithms are tuned to perform best when there is at least 2x swap versus main memory. Configuring too little swap can lead to inefficiencies in the VM page scanning code as well as create issues later on if you add more memory to your machine.