[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: netCDF4 announcement

Rob,

I concur with Mike Folk in welcoming your generous offer to
collaborate on this project and share ideas.

On a related subject, do you think your current parallel netCDF is in
a state that could be disclosed to users who are currently
encountering I/O bottlenecks with netCDF on parallel platforms?  

Even though it's not a 1.0 release version yet, I think there are
others who would like to know about it and might be able to help in
providing feedback about its usefulness.

When you think it's ready for testing/evaluation by some friendly
users, I suggest either announcing it to the
address@hidden mailing list or providing a brief entry
for the "What's New with netCDF?" page at

  http://www.unidata.ucar.edu/packages/netcdf/whatsnew.html

In either case, I'd be happy to assist in getting the news out about
your parallel interface.  I remember getting some very insightful bug
fixes from Jianwei Li last year:

  
http://www.unidata.ucar.edu/cgi-bin/aglimpse/70?query=Jianwei+Li&case=on&errors=0&maxfiles=10

that we have since incorporated into our netCDF 3.5.1beta10 release,
and learning about this project clears up the mystery about how he
found those bugs!

--Russ

_____________________________________________________________________

Russ Rew                                         UCAR Unidata Program
address@hidden                      http://my.unidata.ucar.edu


>Date: Tue, 1 Apr 2003 15:36:37 -0600 (CST)
>From: Rob Ross <address@hidden>
>To: Mike Folk <address@hidden>, <address@hidden>
>cc: Robert Latham <address@hidden>, Rajeev Thakur <address@hidden>,
>    William Gropp <address@hidden>, Alok Choudhary <address@hidden>,
>    Wei-keng Liao <address@hidden>
>Subject: netCDF4 announcement
>
>Mike, all,
>
>We at MCS and NWU are really excited to hear that an effort has been
>funded to enhance netCDF's parallel I/O capabilities and to leverage the
>work on HDF5 in that process [1].
>
>We've been working in a similar direction ourselves to provide a new
>interface more amenable to parallel I/O while retaining the netCDF file
>format.  We have a prototype implementation of this, and preliminary
>performance numbers have been quite good.  We attribute this mainly to
>what we think is a solid interface for parallel access that maintains the
>spirit of the netCDF API while providing the expressiveness necessary to
>leverage parallel I/O semantics underneath.  A web page describing this
>work, including a document describing the API, is available at:
>
>http://www.mcs.anl.gov/parallel-netcdf
>
>We would very much like to contribute to your new effort, if possible,
>rather than duplicating effort.  We have had good success working with the
>HDF5 team before, and we hope that this is another occasion on which we
>can share ideas.
>
>We look forward to hearing back from you!
>
>Rob
>---
>Rob Ross, Mathematics and Computer Science Division, Argonne National Lab
>
>[1] http://www.unidata.ucar.edu/proposals/NASA-AIST-2002/abstract.html