[MITgcm-support] opteron problems, cont'd

Matthew Mazloff mmazloff at MIT.EDU
Fri Jun 1 10:59:49 EDT 2007


Hi Samar,

no idea how many procs you are using or if this is at all related,  
but on bluegene machines i have come across a problem with scratch  
files when using many procs and fixed it by adding

       WRITE(scrname1,'(3a)') 'scratch',myProcessStr(1:4),'_1'
       WRITE(scrname2,'(3a)') 'scratch',myProcessStr(1:4),'_2'

before

       OPEN(UNIT=scrUnit1,FILE=scrname1,STATUS='UNKNOWN')
       OPEN(UNIT=scrUnit2,FILE=scrname2,STATUS='UNKNOWN')

-Matt



On Jun 1, 2007, at 7:45 AM, Samar Khatiwala wrote:

> Hi
>
> I think I figured out part of the problem. If I replace
>
>      OPEN(UNIT=scrUnit1,STATUS='SCRATCH')
>      OPEN(UNIT=scrUnit2,STATUS='SCRATCH')
>
> with
>
>       OPEN(UNIT=scrUnit1,FILE='scratch1',STATUS='UNKNOWN')
>       OPEN(UNIT=scrUnit2,FILE='scratch2',STATUS='UNKNOWN')
>
> in ini_parms.F, open_copy_data_file.F, and eeset_parms.F, the model  
> seems to go much further along.
> This may have something to do with inadequate write permissions in / 
> tmp (or wherever the scratch files
> are created) on the nodes.
>
> However, the model still hangs and now gives the following error:
>
> PGFIO-F-228/namelist read/unit=11/end of file reached without  
> finding group.
> File name = scratch1    formatted, sequential access   record = 10
> In source file packages_boot.f, at line number 1704
> PGFIO-F-228/namelist read/unit=11/end of file reached without  
> finding group.
> File name = scratch1    formatted, sequential access   record = 1
> In source file gmredi_readparms.f, at line number 3554
> Terminated
>
> The above lines correspond to:
>
> READ(UNIT=iUnit,NML=PACKAGES) in packages_boot.F, and
> READ(UNIT=iUnit,NML=GM_PARM01) in gmredi_readparms.F
>
> Any ideas on fixing this would be appreciated. The code runs fine  
> on another cluster (32bit Athlon).
>
> Thanks, Samar
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support




More information about the MITgcm-support mailing list