On Fri, Nov 03, 2006 at 02:55:56PM -0800, Murali Vilayannur wrote:
> Ah I see. So that's how ROMIO works.. Cool then.  My only concern
> was that if we do write buffering, we should not break MPI I/O.  So
> perhaps, we also need to implement a very simple file lock manager,
> fcntl support.  We would also require that ROMIO's ufs driver use
> the fcntl trick if statfs() on the mount point returns pvfs2 sb
> magic number or something..  In fact that brings me to the question,
> does ROMIO figure out if a mount point is a regular ufs mount point
> or a nfs mount point automatically?

Hi Murali. I think for ROMIO we want the file system to get out of the
way. ROMIO can exploit MPI-IO semantics to do reasonable caching on
its own.  Ok, we don't do this now, but it's possible.  As long as
it's easy for ROMIO to turn off any "help" from the underlying file
system, we'll be ok. 

==rob

-- 
Rob Latham
Mathematics and Computer Science Division    A215 0178 EA2D B059 8CDF
Argonne National Lab, IL USA                 B29D F333 664A 4280 315B
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers

Reply via email to