Why use a Pagefile?

[Guest]

As for MCE2005 also in this PC which has 1024MB of physical memory, I have
set Virtual Memory Paging file to off for Vista Ultimate RC 1. There is no
pagefile.sys on harddisk. I was surprised that Vista worked at all since it
needs ~5 times as much physical memory as MCE. It seems to come up somewhat
faster with less disk I/O and even the Media Center works. The question is,
if a PC has >= 1024MB core, why does a pagefile get allocated at all?

 

[Beck]

As for MCE2005 also in this PC which has 1024MB of physical memory, I have
set Virtual Memory Paging file to off for Vista Ultimate RC 1. There is no
pagefile.sys on harddisk. I was surprised that Vista worked at all since
it
needs ~5 times as much physical memory as MCE. It seems to come up
somewhat
faster with less disk I/O and even the Media Center works. The question
is,
if a PC has >= 1024MB core, why does a pagefile get allocated at all?

I have 1Gb ram. When I turned off my page file I started to get program
crashes and memory errors.

 

[Robert Moir]

As for MCE2005 also in this PC which has 1024MB of physical memory, I
have set Virtual Memory Paging file to off for Vista Ultimate RC 1.
There is no pagefile.sys on harddisk. I was surprised that Vista
worked at all since it needs ~5 times as much physical memory as MCE.
It seems to come up somewhat faster with less disk I/O and even the
Media Center works. The question is, if a PC has >= 1024MB core, why
does a pagefile get allocated at all?

Why do you need a pagefile? Because each process running on your machine
thinks it has access to gigabytes of memory address space, and if it tries
to ask for all the space it's entitled to with no pagefile defined you'll
run into problems.

Brief discussion of pros and cons of pagefile use here (talking about XP,
not Vista, but still) :
http://www.codinghorror.com/blog/archives/000422.html

1Gb of RAM is by no means a good limit to be turning off your pagefile on
any OS I would say. I would certainly say that it's far too low a number to
be thinking of this sort of thing in Vista. But it's your computer.

 

[David R. Longnecker]

There are also some applications that specifically look for the existance
of virtual memory and crash with OutOfMemory exceptions if the page file
cannot be found. I have 2GB of memory and tried it a bit back with little
success and no noticable performance increase.

And the blog post from Jeff Atwood that Robert posted is pretty interesting;
the discussion, if nothing else.

-dl

 

[Robert Moir]

There are also some applications that specifically look for the
existance of virtual memory and crash with OutOfMemory exceptions if
the page file cannot be found. I have 2GB of memory and tried it a
bit back with little success and no noticable performance increase.

Good point, I'd forgot about apps actually testing for virtual memory. There
are also some that aren't testing for it as such but which will die without
it due to the way they're coded *sigh*.

Personally, I think running without a pagefile is far more trouble than it's
worth. It might be worth setting the intial pagefile value to a low value
and giving it room to grow, that seems a fair compromise.

 

[Dale \Mad_Murdock\ White]

I guess I've had better results. I've run XP without a Pagefile for months
on end and couldn't see any particular problem. Under the benchmarks I ran,
I didn't see any gains or losses, nor did I see memory faults. I will agree
that 1GB is a bit low to try and run Vista with no pagefile. When I did,
and ran the Half-Life 2 benchmark, I got a Low Virtual memory error. With
2GB, it ran fine with no pagefile.

I currently have my XP running with a 512MB pagefile, and with 2GB of ram, I
have Vista running with a 1.5 GB pagefile. I've played with cranking it up
an down and so far, I don't seee much either way

 

[Robert Moir]

I guess I've had better results. I've run XP without a Pagefile for
months on end and couldn't see any particular problem. Under the
benchmarks I ran, I didn't see any gains or losses, nor did I see
memory faults. I will agree that 1GB is a bit low to try and run
Vista with no pagefile. When I did, and ran the Half-Life 2
benchmark, I got a Low Virtual memory error. With 2GB, it ran fine
with no pagefile.
I currently have my XP running with a 512MB pagefile, and with 2GB of
ram, I have Vista running with a 1.5 GB pagefile. I've played with
cranking it up an down and so far, I don't seee much either way

It depends on the use you put your machine to. I'm just sitting hre reading
email and newsgroups and with 2Gb in my machine it can handle that just fine
if I turn off the pagefile. Many games will work just fine in that memory
space like you say. But I can cause myself no end of trouble if I leave the
system in that state and start working on a big website article which for me
typically involves many browser instances, email, image editing programs,
virtual PC, and maybe visual studio.

 

[Beaker]

It depends on the use you put your machine to. I'm just sitting hre reading
email and newsgroups and with 2Gb in my machine it can handle that just fine
if I turn off the pagefile. Many games will work just fine in that memory
space like you say. But I can cause myself no end of trouble if I leave the
system in that state and start working on a big website article which for me
typically involves many browser instances, email, image editing programs,
virtual PC, and maybe visual studio.

Unless they changed it in later editions...MS Word is one of those
apps that will complain if you have no virtual memory set.

 

[Guest]

What about running the pagefile off on a separate hard drive? Back in the old
days of Win95\98 it sure sped up the computer.

 

[Robert Moir]

What about running the pagefile off on a separate hard drive? Back in
the old days of Win95\98 it sure sped up the computer.

I often tend to do this on a server, but on most desktop machines with
modern hard drives, the difference will be so slight for most people that it
won't make much of a difference. Back in the day it used to make a big
difference to a desktop computer, but that was when no one had much memory
at all, disks were slower, and caching (quite aside from the whole 'less ram
to setup a cache' thing, was considerably less efficient than it is now.

I've got to admit, I'm rather amazed at how many people seem to want to
fiddle with their pagefile. I've seen some tweaks that help servers and
specialised workstation tasks, but on a Win 2000 / XP vintage home machine I
can safely say I've setup machines with tweaked pagefile settings and setup
machines with defaults and not seen much difference one way or another.
Well apart from errors when I turned it all off...

I'd love to know how many people who claim that 'configuring it to value Y
while standing on one leg and whistling 'Start Me Up' by the Rolling Stones
made my machine much faster' have actually conducted real, verifiable and
repeatable benchmarks to prove this (and no, 'd00d, ph33r, I once scored 5
points higher on 3D Mark 2006, it like totally pwn3d' doesn't count).

 

[Dale \Mad_Murdock\ White]

I'll admit I like tweaking for the sake of tweaking. I should also note that
I run my pagefile off a pair of drives running in RAID 0, for that extra
..0001% performance boost.

Obviously, I think to quailfy your standard of test, I'd have to run a
mahine with say around 128-256 Meg of ram. Obviously, the more apps I use,
the more page file I will hit. I guess the hardest question is how do you
really measure the performance gains.

My only semi real world numbers that I would know how to do it, Would be to
run 128-256 MB of ram, then run a game like FEAR with the different pagefile
settings and see if I can translate it into something noticable. It's easy
to notice during game play, because it will stutter enough to drive a person
mad, but I haven't always seem that translate into really low numbers on
canned benchmarks.

Plus, I normally spend that exercise to prove the vaule of adding more ram
and eliminating the need for a pagefile at all (or at least, drastically
reducing the actual usage of it). I'll agree with the notion of, why bother
tweaking a pagefile, now that RAM itself is so cheap (compared to 10 years
ago)

I'd say the only reason I tweak the pagefile, which really isn't a tweak, is
that I don't need a 3GB pagefile on my system. I have 2GB of ram, so I
really don't see the need to go with XP's 1.5x pagefile increase. Though I
have ample disk space. Even a 2GB file just seems wasteful.

 

[Robert Moir]

I'll admit I like tweaking for the sake of tweaking. I should also
note that I run my pagefile off a pair of drives running in RAID 0,
for that extra .0001% performance boost.

There's nothing wrong with either of those things of course, I've been known
to do the same myself. As long as you know why you're doing something
instead of just following a vague idea on a list, there's nothing wrong with
playing with any part of your system that takes your fancy ;-)

 

[Rock]

I'd love to know how many people who claim that 'configuring it to value Y
while standing on one leg and whistling 'Start Me Up' by the Rolling
Stones made my machine much faster' have actually conducted real,
verifiable and repeatable benchmarks to prove this (and no, 'd00d, ph33r,
I once scored 5 points higher on 3D Mark 2006, it like totally pwn3d'
doesn't count).

Good song...lol. It's amazing he's still going strong.

 

[Homer J. Simpson]

I'll admit I like tweaking for the sake of tweaking. I should also note

that I run my pagefile off a pair of drives running in RAID 0, for that
extra .0001% performance boost.

Wouldn't RAID-1 make a more noticeable difference, speed-wise?

 

[Robert Moir]

Wouldn't RAID-1 make a more noticeable difference, speed-wise?

Yes but in the wrong direction. If your RAID 1 setups are faster than your
RAID 0 setups then you need new disks or a new RAID controller or both.

RAID 0 = Striping = fast but unreliable
RAID 1 = Mirroring = roughly normal disk performance (possibly faster
reads), wasteful of capacity, but reliable.
RAID 5 = Stripe with parity = often lower performance, but very reliable and
good ROI on disk hardware.

 

[Dale \Mad_Murdock\ White]

I agree, RAID 0 is suppose to be the best (or at least the fastest) in terms
of both read and write speeds, mostly because it's "suppose" to read and
write to 2 disks at the same time (or near same time). RAID 1, should be
equally fast on reads, since again it's suppose to read from both disks at
the same time, writes re a tad slower on raid 1

Of course, this is making certain assumptions. Normally the really expense
high end raid systems give you the real performance gains. The $100 home
RAID card(s) don't seem to really off all that much of a performance gain.
But even that is subjective, because it depends what you're doing with it.
In my case, teh RAID 0 drive is a scratch volume for all things, whether
playing with video encoding, ripping CDs to MP3, etc etc. In this case, I
started out wanting maxing space with no concern to data loss. It's the
problem with having a 12 bay case, It just doesn't seem right if I don't
have at least 10 drives in it.

 

[Homer J. Simpson]

Wouldn't RAID-1 make a more noticeable difference, speed-wise?

Yes but in the wrong direction. If your RAID 1 setups are faster than your
RAID 0 setups then you need new disks or a new RAID controller or both.

RAID 0 = Striping = fast but unreliable
RAID 1 = Mirroring = roughly normal disk performance (possibly faster
reads), wasteful of capacity, but reliable.
RAID 5 = Stripe with parity = often lower performance, but very reliable
and good ROI on disk hardware.

You're right, I got those mixed up (pre-coffee post syndrome)

FWIW, performance-wise, I had a RAIDed (RAID-1) Athlon64 3200 that booted
*very* quickly (too bad I never timed it). I replaced the CPU for a 4800,
but had to get rid of the RAID configuration (same drives though).
Result--the machine now *easily* takes 3 times as long to boot up, despite
the faster CPU.

 

[Dale \Mad_Murdock\ White]

Just an Interesting side note. First time every, I've ran into a video game
that won't play without a page file. Not only does it want a pagefile, but
it demanded a pagefile of at least 768MB. In this case, I was running a page
file of 512MB.

Which is interesting is that I'm running 2GB of ram and this was on XP.
Looking at perfmon at the end of a few mission, total ram usuage was 1.3 GB,
so on paper it shouldn't have needed that pagefile.

Anyways, the game was the Company of Heros SP Demo. Just thought it was
interesting that it demanded a 768 MB pagefile or a pagefile in general

 

[Phillips]

You can measure read/write hard-drive speeds w/ any of many tools available
free on the web; Raid0 does make a noticeable difference of at least 50%
faster for 2 drives array. I've been using Raid0 w/out any problems for 2
years now... only when a hard-drive dies (mechanically) you lose whatever's
on the array. I wouldn't even think of not running the OS from a partition
on the Raid0 array - with images of it just in case.
Michael