Opened 6 years ago

Closed 6 years ago

#1589 closed Bug (fixed)

AGRIF not running with XIOS

Reported by: nicolasmartin Owned by: nemo
Priority: low Milestone:
Component: AGRIF Version: trunk
Severity: Keywords:
Cc: Branch review:
MP ready?: Task progress:

Description

As I work to add an AGRIF Agulhas configuration to be checked by NEMO trusting, I'm not able to run this config with XIOS, either on ADA or CURIE HPC.

On ADA with 32 procs, opa crashed during initialization between ssh_swp and dia_obs in ocean.output and just after agrif_nemo_init in 1_ocean.output. I get the following message from ADA monitoring:

> Error [void noMemory(void)] : In file '/workgpfs/rech/omr/romr005/XIOS/xios-1.0/src/memory.cpp', line 9 -> Out of memory

> Error [void noMemory(void)] : In file '/workgpfs/rech/omr/romr005/XIOS/xios-1.0/src/memory.cpp', line 9 -> Out of memory

> Error [void noMemory(void)] : In file '/workgpfs/rech/omr/romr005/XIOS/xios-1.0/src/memory.cpp', line 9 -> Out of memory

ERROR: 0031-250  task 14: Aborted
ERROR: 0031-250  task 22: Aborted
ERROR: 0031-250  task 7: Aborted
Insufficient memory to allocate Fortran RTL message buffer, message #78 = hex 0000004e.

On Curie with 128 procs, same lines in *ocean.output files but opa didn't crash and seemed to work in a loop in the initialization with no time step during 10 mns.

I ran simultaneously the same namelists with an opa executable compiled without XIOS and the jobs reached the expected 150 time steps, either on ADA or CURIE.

Commit History (0)

(No commits)

Change History (1)

comment:1 Changed 6 years ago by nicolasmartin

  • Resolution set to fixed
  • Status changed from new to closed

By reducing the amount of procs (/2), the runs completed.

I guess I'm experiencing the joint problem of buffer size requested against the computing architecture limitations and the reason why XIOS could be run in detached mode.

Note: See TracTickets for help on using tickets.