[mpich-discuss] ADIOI_GEN_WriteStrided(358): Unable to write to the file as part of a strided write operation

[Rob Latham]

On Thu, Dec 22, 2011 at 01:21:16PM -0800, burlen wrote:
> Hi,
> 
> Can anyone clarify/insight into the error message? What could cause
> it? How to fix it?

Huh.  'ADIOI_LUSTRE_IOCONTIG' suggests it's a lustre thing.   To get
that message, you have ended up in the simple "read or write this
contiguous hunk of data" function. 

It looks like the strerror() version of whatever went wrong was
"Unable to write to the file as part of a strided write operation".  

Google isn't any help as to what that message means for lustre. 

You could try setting the MPI-IO hint "direct_write" to "true", but
mostly I'm just curious if that does anything.  

Is there some way to enable collective I/O in SciVisToolKit ?  Setting the
MPI-IO hint "cb_write" to "enable" might help, too.  That
would reduce contention to the file and may make this error go away.

It's possible that Lustre is returning an interrupted system call and
that ROMIO should handle it by retrying the operation.  

It's a long shot but maybe the folks on the lustre mailing list will
know what's going on?  

Sorry, never seen this error before.  Um.. I'm about to head off for
winter vacation, but I'll check back with you in the new year.
Perhaps santa will bring me some more ideas of things we can do to get
you back on track.


==rob



> [158] Error in:
> /home/01237/bloring/ParaView/SciVisToolKit/release/SciVisToolKit-11.02/MPIRawArrayIO.hxx,
> line 594
> Error writing file.
> Other I/O error , error stack:
> ADIOI_GEN_WriteStrided(358): Unable to write to the file as part of
> a strided write operation
> ADIOI_LUSTRE_IOCONTIG(164).: Other I/O error Input/output error
> Error in:
> /home/01237/bloring/ParaView/SciVisToolKit/release/SciVisToolKit-11.02/BOVWriter.cxx,
> line 255
> WriteDataArray ue component 0 failed.
> _______________________________________________
> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> To manage subscription options or unsubscribe:
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss

-- 
Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA