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

Re: parallel implementation of NetCDF ?



> Organization: University of Maryland at College Park
> Keywords: 199405031954.AA23337

Hi Francois,

> I am wondering if anybody made (or is planing on making) a parallel
> implementation of NetCDF on machines like Intel Paragon, TMC CM-5,
> Cray T3D, etc. which would take advantage of available parallel IO on
> the target machine.

Yes, Rick Light from Los Alamos (address@hidden) was very close to
completing a port of netCDF to a CM-5 and had even managed to exploit some
parallelism, if I remember correctly from the last time I talked to him
about this.

Also, we have had contact from programmers from Cray Computer Corporation
about porting netCDF to the Cray 3 in a way that exploited vectorization for
floating-point reads and writes.  I know Chris Anderson
(address@hidden) from NERSC has some interest in porting netCDF to a
T3D.  Tim Sheehan (address@hidden) from ORNL has tried to port
netCDF to an Intel Paragon though I don't know if he succeeded.  Harry
Moffat (address@hidden) from Sandia Labs was also interested in
getting netCDF ported to an nCube and a Paragon.  Mike Wehner from LLNL
(address@hidden) was working on porting netCDF to a Paragon last
fall, but I think he ran into problems with Intels XDR library, and I don't
know whether he solved them.

> I have posted this question to the Newsgroup sci.data.formats and
> received responses indicating a strong interest in the field. Also, I
> attended the DOE/CHAMMP (global climate change) meeting last week at
> Albuquerque and there was also a strong interest in having a parallel
> implementation of NetCDF (the NCAR's CCM2 will be using NetCDF soon).

We aren't currently planning on developing our own port to any parallel
machines because we don't have convenient access to any parallel machines,
and it's not in Unidata's current mission to develop software for such
platforms since none of our sites are now using parallel machines.  However,
we may soon find a need to exploit multithreading on SPARCstations and other
workstations with multiple CPUs, so we may eventually have to addreess the
issues.

If someone else tackled this problem and made modified netCDF sources
available, we would be willing to help point others to them and add them to
our archive of contributed software.

--Russ
__________________________________________________________________________
                      
Russ Rew                                              UCAR Unidata Program
address@hidden                                        P.O. Box 3000
(303)497-8645                                 Boulder, Colorado 80307-3000