Intel Fortran Compiler 9.1 and 10.1 compiles extremely slow on newest RedHat EL 5

[urte-fuerst]

Hello,

we experience very slow compilation progress on the newest RedHat EL5 Linux. This occurs with Intel FORTRAN compilers at least in version l_fc_c_9.1.039 and l_fc_p_10.1.015. The linux architecture is x86_64.
On a slightly older RedHat EL4 (also x86_64) the performance is much better:
For a bunch of FORTRAN 117 source files, the compilation with ifort -c -g *.f takes 5 seconds on the EL4 machine and 63 seconds on the EL5 machine. It looks like there is some environment processing done on the EL5 (sourcing shell startup scripts), which I don's see on the EL4.

Has anybody else experienced something similiar? Any ideas how to speed-up things on RedHat EL5?

Best regards,
Urte

[TimP]

By "newest," do you mean update 2? Virtual kernel or not?

[urte-fuerst]

Yes, update 2. No virtual kernel.

[Kevin_D_Intel]

Urte-Fuerst:

It looks like there is some environment processing done on the EL5 (sourcing shell startup scripts), which I don's see on the EL4.

What additional details can you share about what led you to suspect this?

[urte-fuerst]

Because at our site we have a site-wide stdbashrc | stdcshrc for common environment settings which gets sourced when opening a new shell. This issues an information about the current DISPLAY setting like this:
display: :0.0
When compiling on EL5, I see this "display: :0.0" coming on stdout after issuing the ifort command. I don't see it when compiling on EL4. That's why I suspect some additional (unnecessary?) shell startup sourcing going on. I doubt that it slows down compilation enough to explain the effect, but it's the only hint I have a the moment.

[Kevin_D_Intel]

ifort is a shell script itself that execs the real ifortbin compiler executable. If your site-wide stdbashrc/stdcshrc are sourced when a new shell spawns then those are perhaps at play here when invoking ifort.

You could try invoking ifortbin directly with same command options used with ifort to see if there is any change to the display diagnostic on stdout.

It is curious why you do not see the same display diagnostic on your EL4 system is if the same site-wide files are used on that system too.

[urte-fuerst]

Well, when I wanted to make this test, the slow compilation effect had simply gone away on RedHat EL5. ?? We observed it for about at least 2 weeks and it stayed there even after rebooting the machine. As it looks now it doesn't seem to be really an intel compiler problem but some network-related/NFS-related problem with RedHat EL5.

This means the problem remains a very strange one (because the RedHat EL4 boxes of course mount from the same software server as the RedHat EL5 without having the same difficulties), but this forum is presumably not the place to discuss it. Might be something to discuss with RedHat...

Anyway thanks for your time and your effort!
Best regards,
Urte