[lustre-discuss] Problem with standard fortran IO and lustre fs, sporadically slow io rate

[Andreas Dilger]

On Feb 3, 2022, at 05:13, Bertini, Denis Dr. <D.Bertini at gsi.de<mailto:D.Bertini at gsi.de>> wrote:

I am facing a problem when trying to write on a Lustre filesystem ( v 2.12) using standard fortran IO
code ( compiled with GNU gfortran 8.3).
The io rate sporadically goes extremely slow ( sometime down to  20/10 kBytes/s ! ) giving the
feeling that the job is just hanging.
Is this problem known by the community?
Is there something a user should be aware when using Fortran IO together with
the lustre filesystem?

The code (tarball in attachment) i am using is very simple and i would be grateful is somebody could try it to see if this behaviour is a local or general problem.

It's been 30 years since I did anything directly with Fortran, so I can't help you there, but I would suggest to debug this from the IO side with tracing.  Use "strace" if a single-threaded program, or Darshan if a parallel program, or other related tools, and you can get stats about IO size and alignment without having to understand the code first.  This may surprise you if the application is doing e.g. 32-byte writes because you specified the array variables in the wrong order, or something.

Cheers, Andreas
--
Andreas Dilger
Lustre Principal Architect
Whamcloud







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20220204/dd6a60ac/attachment.html>