[CIG-SEISMO] SPECFEM3D GLOBE adios_group_size warning

[Matthieu Lefebvre]

Hi Dirk,

There is parameters to tune ADIOS behavior; in particular you can try a different buffer size.
https://github.com/geodynamics/specfem3d_globe/blob/master/setup/constants.h.in#L349

You will need to reconfigure after changing any of the parameters in this file.

Best,
Matthieu


From: CIG-SEISMO <cig-seismo-bounces at geodynamics.org> on behalf of van Herwaarden Dirk Philippus <dirkva at student.ethz.ch>
Reply-To: "cig-seismo at geodynamics.org" <cig-seismo at geodynamics.org>
Date: Wednesday, May 25, 2016 at 10:58 AM
To: "cig-seismo at geodynamics.org" <cig-seismo at geodynamics.org>
Subject: [CIG-SEISMO] SPECFEM3D GLOBE adios_group_size warning

Dear all,

I am using SPECFEM3D Globe in GPU_Mode to perform a full waveform inversion of Africa for
a MSc thesis project. I am performing a regional simulation using one big chunk centered around Africa on the Piz Daint supercomputer. However when I try to increase the amount of elements from 256 to 384,  I run into problems. The amount of node hours required for 1 simulation increases very fast (It takes about 20 times more time) and my .stderr file shows many of these warnings:

"WARN : adios_group_size (SPECFEM3D_GLOBE_FORWARD_ARRAYS     1): Not buffering. needs: 460596347 available: 209715200."

I felt that these warnings might be the reason why the amount of node hours required for one solve increases so much. Am I making a mistake or is there a parameter that I can change to prevent this from happening? I only started working with SPECFEM recently.
Any help would be highly appreciated!

Best regards,

Dirk van Herwaarden

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/cig-seismo/attachments/20160525/4bcef60b/attachment.html>