eXP Pagefile size hell!

  • Thread starter Thread starter Alex Thalmann
  • Start date Start date
A

Alex Thalmann

Hi all, hope someone can help :)

I'm unable to control the size of my eXP pagefile using the Target
Designer (or via registry entries). I'm using Perfmon to monitor the
performance of the embedded system, so I try, as stated in other posts
and info on the web, to set the pagefile size to min=2mb and max=2mb.
However, eXP seems to ignore these settings, because after the
installation, the pagefile is some random size, usually larger than
64mb. I've confirmed the registry entry in the device's registry reads
'pagefile 2 2'. I've tried other sizes, but with anything less than a
fixed 16mb pagefile, the OS sets its own file size.

My installation size is around 110MB uncompressed, which will
hopefully be compressed (around 70mb) onto a 128mb CF card (if I can
get a 2mb pagefile!).
The system I am using has 128mb of ram and currently, the partition
onto which the OS is installed is 190MB with the same value set in TD
(which will change later for the CF).

Any ideas?
 
Have you tried to disable the page file? at 2mb it is not giving you much
in the first place.

Also, you will probably not want a page file on your CF card.
 
Dear Rick,

Alex wants to use "Performance Monitor", therefore he must sets the
pagefile.

However, I have no idea to limit pagefile size.


Best Regards,
Bing Chen
2004.05.28
 
Bing,

Good point. Missed the "Performance Monitor" , Funny how the mind skips
over things
 
I'm not sure if this is related, but I have a similar problem in
setting the EWF size. No matter what size i set in TD, it sets its own
size, usually 64MB.

I'm using an evaluation version of the eXP development environment
with the latest patches for EWF.
 
Alex,

This is an issue with XP (and therefore XPe). If you set a page file with a
size < 16Mb on any partition, on next boot XP kernel will create the
pagefile on the system (!) partiton with the "recommended" size ~1.5 of RAM.
The funny thing is that before the reboot the System Control Panel applet
will create the pagefile of specified size (e.g., 2Mb) on the specified
partition (not necessarily system one), but the kernel switches to the
system partition (and 1.5 RAM size) pagefile on the next boot.
I've been observing this "undocumented" (if not buggy) behaviour with XPe
and XP Pro. Win2K kernel does not seem to have the problem (you can create
and keep pagefile with any size >= 2Mb).

I don't know how to fix the problem in XP kernel.

KM
 
Doesn't matter why you want to set a page file - it will trash the CF just
the same.

S
 
Hi Simon,

I'm just wondering if someone tried to use pagefile on partition protected by RAM EWF?
Does it go around FS like hibernation support or trough FS.

Regards,
Slobodan
 
Not at all sure, and not tried it. The original post didn't mention EWF -
Alex Tlalmann later said that EWF (initial overlay size?) exhibited similar
behaviour.

Strikes me as a bit desparate (not to say recursive) to need to put a paging
file into a RAM overlay.

IMHO it would be better to swallow hard and find a better and more embedded
friendly way of getting the performance stats

Rest regards

Simon
 
Strikes me as a bit desparate (not to say recursive) to need to put a paging
file into a RAM overlay.
Click to expand...

I absolutely agree on that.
IMHO it would be better to swallow hard and find a better and more embedded
friendly way of getting the performance stats
Click to expand...

More embedded way would be not to use them in the first place.

Regards,
Slobodan
 
When I tested the pagefile functionality with RAM EWF to find out the limits, the 1.5GB pagefile was created on the protected volume
without EWF usage going up (I had only 1G RAM).
 
Sorry, I meant EWF RAM usage.

Basically, I have a protected volume with RAM EWF enabled on it. The volume contains an active pagefile.sys (~ 1.5GB).
I don't think FS is used to create the pagefile.
 
Hi Konstantin,

I know what you meant, although part of this sound strange.

For instance if page file was 64 MB and then it grow to 512 MB, filesystem must be aware of that change in size.
Content access is another story and it probably goes directly to disk driver like we saw that happens with hibernation file.

But then this mean that idea of using EWF with small pagefile would not give any results.
Perhaps some form of ramdisk driver would do the job then. If we put pagefile on ramdisk. Just a thought.

Best regards,
Slobodan
 
Slobodan,
For instance if page file was 64 MB and then it grow to 512 MB, filesystem
Click to expand...
must be aware of that change in size.

What I saw (I must probably test it more carefully since I did not pay
enough attantion to the pagefile with EWF on) is that the pagefile does not
get created in EWF is on (most likely it gets created with
CreateFile/ZwCreateFile/IoCreateFile API).
Again, the same as with hiberfil.sys.
Content access is another story and it probably goes directly to disk
Click to expand...
driver like we saw that happens with hibernation file.

Yup, that's what I thought. Otherwise I don't understand why EWF does not
eat all memory.
However, I think I have not tested that enough. I would need to make sure
all the memory is not paged or to allocate much memory to force the system
to use the pagefile. Then I may be sure that EWF is not passed by.

I'll do more testing tomorrow (don't have a Minlogon image on the hands
now).
But then this mean that idea of using EWF with small pagefile would not
Click to expand...
give any results.

A small pagefile is almost useless anyway. Why would one need to use EWF
with a small pagefile?
What results do you expect?
Perhaps some form of ramdisk driver would do the job then. If we put
Click to expand...
pagefile on ramdisk. Just a thought.

Then you double the RAM usage for any alloc (compared to not having pagefile
at all), don't you? And you don't gain anything by having pagefile on
ramdisk. E.g.,
RAM=500MB, OS=100MB, Ramdisk=100MB.
Without pagefile and ramdisk, I am able to alloc up to (RAM-OS=400MB).
With pagefile on the ramdisk, I am able to alloc up to
(RAM-OS-Ramdisk+Pagefile=400MB).

Regards,
Konstantin
 
Hi KM

Thanks for the input.

I have come to the same conclusions, ie: this is an XP issue, since
the same thing happens on a number of other PC's I have running XP
Pro.

As for the the requirement of the pagefile, from my testing with a
fixed pagefile of 16MB and our application, the EWF RAM usage reaches
no more than 2MB after the initial 'settling down' period, so for our
use, using the pagefile protect by the EWF would be ok. From what I've
seen using Perfmon, no diskwrites occur on the partition with the
pagefile.

Its a tough call, but it looks like we will have to forego the use of
the pagefile (my personal preference anyway).

*Rant* eXP development feels way too 'ragged', but I suppose one
should expect that from an OS that isn't truly componentized to begin
with. *Sigh* :)

Thanks for all the feedback.
 
Konstantin,

And you don't gain anything by having pagefile on ramdisk. E.g.,
You gain pagefile support. That someone wants for performance counting.
I do not use pagefile since I don't need it for any reason.

Best regards,
Slobodan
 
Slobodan,

Yup, only performance counting. I forgot someone may need it :-)

I am still unclear on how the paging is implemented. Why it bypasses EWF.
What I see is that RAM usage does not come up even if I have a lot of page faults (pagefile is used).
With a test app I consumed way more than Available physical memory, and Virtual/Pagefile memory was certainly used. However, I was
not observing the additional RAM usage coming from EWF. Therefore, I assumed paging was bypassing EWF.
Even if I had a small pagefile and after some allocations got "Low Memory.. adjusting the pagefile size.." system message and the
pagefile was increased to the maximum specified, there was no EWF RAM usage increase.

I thought (guessed and some googling helped) that MM paging writes are implemented with calls to IoSynchronousPageWrite
(IRP_PAGING_IO). But then they definitely should end up calling to FS driver and then to EWF driver.

This is probably a question to comp.os.ms-windows.programmer.nt.kernel-mode. Or to whoever created EWF.

Regards,
Konstantin
 
Back
Top