Gaussian 03 & PGC & Fedora core 4.0

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hi,
I got this error message once Gaussian g03 is compiled on fedora core 4:
Erroneous write during file extend. write 208 instead of 4096
Probably out of disk space.
Write error in NtrExt1: No such file or directory

And I do not have any problem of space… This problem has been already reported @

but I did not find the answer… Is the answer known ?

Thanks, Francois

2

Hi Francois,

We have not heard of a solution to this problem (nor the cause). I’ve asked our application engineer who works with Gaussian to contact them and ask if they have seen this issue before. Hopefully, we’ll have more information later.

Thanks,
Mat

3

Hi Mat,

OK thanks for your answer… I am going to do more tests: I let you know what I get…

  • First the same problem appends with g98 & FC 4.0.
  • Second g98 and FC 2.0 works fine…

Regards, Francois

4

Hi Mat,

I think it is a kernel problem… I compiled on my FC 4.0 system the version 2.6.12 and used it instead of the default 2.6.11 and:

  • g98 was able to work this time (with 2.6.11 It did not work)
  • g03 jobs still stop, but a little latter (in the output)…

I will try other kernels. I let you know. regards, Francois

5

Hi Francois,

Our contact at Gaussian sent the following reply:

This is a strange error, because it suggests that part of a write
completed but not the rest, yet the error is “no such file” which
suggests that a scratch file has been deleted. Could this have
been using an NFS mounted file system for GAUSS_SCRDIR and the server
went away?

The other possibility is that there is some problem with the calls we
make to do i/o to large files. I think there is some indication that
RedHat messed with this, although the other problems I know about
involve running binaries built elsewhere on some recent RedHat
versions. We don’t support Fedora, by the way.

Sounds to me that its most likely an incompatability with Fedora Core 4. However, you might want to contact Gaussian directly since they’ll should be able to help you better with this problem.

Thanks,
Mat

6

Mat,

This is a strange error, because it suggests that part of a write
completed but not the rest, yet the error is “no such file” which
suggests that a scratch file has been deleted. Could this have
been using an NFS mounted file system for GAUSS_SCRDIR and the server
went away?

This is obviously not a NFS problem since I do not have any NFS server set up on my machine…

The other possibility is that there is some problem with the calls we
make to do i/o to large files. I think there is some indication that
RedHat messed with this, although the other problems I know about
involve running binaries built elsewhere on some recent RedHat
versions.

I think this is a kernel problem…

I already contacted Gaussian & they told me:
Downgrade to the 2.6.10 kernel & we do not support FC…

I am now compiling kernel 2.6.12.2 and working with 2.6.10… I let you know…
Francois

7

Run

echo 0 > /proc/sys/kernel/randomize_va_space

and put this command line in /etc/rc.local