buffering writes

[Foucar, James G]

Hi Rob, thanks for the response. I have a related question: do you have any
idea how long it's going to be until the nonblocking pnetcdf operations
(ncmpi_iput_vars) become available?
Thanks!
James

-----Original Message-----
From: Rob Ross
To: Foucar, James G
Cc: 'parallel-netcdf at mcs.anl.gov'
Sent: 4/9/2004 7:29 PM
Subject: Re: buffering writes

Hi James,

This is a very good thing to be thinking about.  Actually though, there 
are no MPI-IO implementations that really do this sort of thing right
now, 
and we haven't built this into PnetCDF either.  It's a good idea, and I 
expect to see more implementations (both MPI-IO and high-level libraries

such as PnetCDF) doing this sort of thing in the future.

Rob

On Fri, 9 Apr 2004, Foucar, James G wrote:

> Hello,
> 
> I have a question about buffering in pnetcdf. I am in a unique
situation in
> that the reads and writes in my program are in two different phases.
> Therefore, all of the writes, none of which overwrite each other, do
not
> need to be "seen" until I close the dataset. I was thinking it would
be nice
> to buffer as many of these writes as possible since there is no
urgency in
> them reaching the disk. The problem is I don't know how to enable
maximum
> buffering. Does it have something to do with the MPI_Info object that
you
> give the ncmpi_create procedure?
> 
> Thanks,
> James
> 
>