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

[netCDF #NOJ-562667]: NetCDF Performance

This archive contains answers to questions sent to Unidata support through mid-2025. Note that the archive is no longer being updated. We provide the archive for reference; many of the answers presented here remain technically correct, even if somewhat outdated. For the most up-to-date information on the use of NSF Unidata software and data services, please consult the Software Documentation first.

  • Subject: [netCDF #NOJ-562667]: NetCDF Performance
  • Date: Sun, 19 Aug 2007 20:08:43 -0600
> Russ,
>
> I agree that something looks fishy.
>
> Unless the buffers are huge ...   Would they be hundreds of megabytes?
> The third and sixth block of writes are writing more than 200 MB of data
> in each case.  It is easy to change the clock resolution.  Change the
> definition of clockRate near line 41.  Make it, say 1.e6/ticksPerSec and
> change the time units in the format statements from ms to microsec or
> s.  From the second line of output, the resolution of your clock seems
> to be 1 s.
>
> I would guess that your workstation is much faster than the individual
> processors on most HPC machines.   If it also has huge buffers or clever
> compiling, it might just move a pointer or an array for something that
> looks like I/O to me.
>
> Lee
>
>

Howdy Lee!

I can't shed any light on your specific questions, but I have just added your
program (under the name tst_io.f90) to the netCDF snapshot, and it is run if
the --enable-benchmarks option is used with configure.

We are starting some work on benchmarking some various things with netCDF and
netCDF-4, and will have some results on our website in the coming months.

Thanks!

Ed

Ticket Details
===================
Ticket ID: NOJ-562667
Department: Support netCDF
Priority: Critical
Status: Closed