parallel netcdf / hdf

[John Tannahill]

Richard,

Thanks for continuing to pursue this; I probably won't get to it
today.  I will soon have a pretty good set of parallel netcdf
timing numbers for seaborg (NERSC IBM-SP), mcr, and tckk
(different number of processors, different file sizes, parallel
versus serial, etc.).  While I would very much like to have some
hdf5 timings from at least one of the machines (probably seaborg)
to compare with, it is not a show stopper.  Will need to start
writing things up next week and make sure things are complete by
the end of the month.

By the way, my test codes are set up to mimic as closely as
possible how we are currently doing serial netcdf I/O in our major
model, as well as how we would do something comparable with
parallel netcdf.

Regards,
John

Richard Hedges wrote:
>> Richard,
>>
>> I will take whatever you have with regards to translating my
>> parallel netcdf test code to hdf5.  My time available to work
>> on this stuff is running out.
>>
>> Thanks,
>> John
>>
> 
> 
> Hi John,
> 
> It looks to me like the netcdf machinery is better set up for the access 
> pattern that you are using in your test code.  For HDF5, there is a 
> concept called "hyperslab", which is like the access pattern that you 
> are using.
> 
> see: http://hdf.ncsa.uiuc.edu/HDF5/doc/Tutor/phype.html
> 
> 
> 
> 
> 


-- 
============================
John R. Tannahill
Lawrence Livermore Nat. Lab.
P.O. Box 808, M/S L-103
Livermore, CA  94551
925-423-3514
Fax:  925-423-4908
============================