Paging IO with large files

[Villy Madsen]

I don't know where to go with this, perhaps someone can point me in the
right direction.

To put it in perspective:

Systems affected: WIN/XP - started happening about summer of 06

Environment: Win/XP SP2 patched up
The hardware is a P4 1.5 ghz with 1GB of real and 1.6GB of paging space
allocated (of 4 GB).
1 physical drive with 3 partitions of 36GB, 98GB and 100GB

Application: A VAX emulator - mounting two virtual disks (windows files) of
2GB and 4GB respectively (happened on smaller files as well but I can't
quantify that)

Compilers - various i/c GCC running under minGW and Microsoft C++

Systems - not always consistent

The real system (not the virtual vax) starts writing like crazy - one byte
for every byte on the virtual disk - and the page faults start climbing like
crazy. - it appears to be one page fault for every 4K bytes witten... Bytes
read are low

example for a "VAX" that's been running for a couple of hours

cpu - 10:20 ; MemUsage 15Mb ; Pk Mem 93mb; page faults 1,561,136; VM Size
532 MB;
I/O Reads 692K; I/O writes : 98K; I/O Read Bytes 51Meg; I/O Write Bytes:
6.2GBytes

While the "massive IO" is taking place, the virtual machine is hung. It
starts happening when an IO event is executed by the virtual machine SCP.
This io event could be a write, or a close or perhaps a read.

It's not consistent in when it happens - but the most frequent and readily
traceable event is when the virtual machine is shutting down. The "system
shutdown" message is received on the VAX "console", the SCP closes the file
that is the virtual disk and then windows goes crazy with the IO.

I have looked at the process with process Explorer and NO files are memory
mapped. I do not use disk compression and indexing is turned off. I use
Symantec anti-virus, but the file extensions for the virtual disks are
excluded. (also note - it DOES NOT happen when the files are opened. (I am
even reasonably sure that it does not happen with the first read))

Additional information: It started happening suddenly _ Aug or Sep of 06.
I tried running the "VAX system" on a laptop and it took forever to
startup... It COULD have been coincidental with the install of SP2. The
laptop is a "managed" system. I had used it on the laptop previously
without noticing this problem.

I am not the only one with the "problem". It never happened with w/98.

Anyone have any ideas - or a suggestion as to where I should go for
suggestions ??

Thanks

Villy

 

[Jim]

I don't know where to go with this, perhaps someone can point me in the
right direction.

To put it in perspective:

Systems affected: WIN/XP - started happening about summer of 06

Environment: Win/XP SP2 patched up
The hardware is a P4 1.5 ghz with 1GB of real and 1.6GB of paging space
allocated (of 4 GB).
1 physical drive with 3 partitions of 36GB, 98GB and 100GB

Application: A VAX emulator - mounting two virtual disks (windows files)
of
2GB and 4GB respectively (happened on smaller files as well but I can't
quantify that)

Compilers - various i/c GCC running under minGW and Microsoft C++

Systems - not always consistent

The real system (not the virtual vax) starts writing like crazy - one byte
for every byte on the virtual disk - and the page faults start climbing
like
crazy. - it appears to be one page fault for every 4K bytes witten...
Bytes
read are low

example for a "VAX" that's been running for a couple of hours

cpu - 10:20 ; MemUsage 15Mb ; Pk Mem 93mb; page faults 1,561,136; VM Size
532 MB;
I/O Reads 692K; I/O writes : 98K; I/O Read Bytes 51Meg; I/O Write Bytes:
6.2GBytes

While the "massive IO" is taking place, the virtual machine is hung. It
starts happening when an IO event is executed by the virtual machine SCP.
This io event could be a write, or a close or perhaps a read.

It's not consistent in when it happens - but the most frequent and readily
traceable event is when the virtual machine is shutting down. The "system
shutdown" message is received on the VAX "console", the SCP closes the
file
that is the virtual disk and then windows goes crazy with the IO.

I have looked at the process with process Explorer and NO files are memory
mapped. I do not use disk compression and indexing is turned off. I use
Symantec anti-virus, but the file extensions for the virtual disks are
excluded. (also note - it DOES NOT happen when the files are opened. (I am
even reasonably sure that it does not happen with the first read))

Additional information: It started happening suddenly _ Aug or Sep of 06.
I tried running the "VAX system" on a laptop and it took forever to
startup... It COULD have been coincidental with the install of SP2. The
laptop is a "managed" system. I had used it on the laptop previously
without noticing this problem.

I am not the only one with the "problem". It never happened with w/98.

Anyone have any ideas - or a suggestion as to where I should go for
suggestions ??

Thanks

Villy

While I do not know how the VAX emulator works, on a real VAX the large
number of page faults indicates that the pagefile is way too small. This
large number of page faults consumes so much IO that the program seems to be
hung (although it isn't).

You might post this problem to comp.os.vms which is where the real VMS
experts hang out. At least you would be much more likely to find someone
who is using the VAX emulator (actually, it impersonates VMS).

Jim

 

[villy]

Jim

The problem isn't with the VAX emulator - the problem is what is happening
with the real machine (WinXP).
The most frequent occurence is that the excessive paging (which occurs only
on the real box, the virtual system hardly pages at all) occurs after the
virtual server has shut down and the emulator (the SCP running under
windows) closes the file(s) that are the virtual disk.....

Villy