[Parvis-users] ParNcl and GRIB2 file

[David Brown]

2- Is the graphics production time (For big maps with a resolution of 16384
x 16384 pixels) is improved with ParNCL

Regarding the graphics, the NCL team has made considerable progress
speeding up the raster graphic output using OpenMP.  This is not yet
integrated into the ParNCL code, but should be sometime this fall. The
highest resolution plots we have attempted so far have been in the order of
8000x8000 pixels, partly because we do not have data with high enough
resolution to warrant more pixels. The
8000x8000 plots were used to render an MPAS grid with 65 million cells.

Given this data, the timings do seem to be significantly dependent on the
output resolution. I just tried creating output at
16384 pixel resolution and find the time more than doubles compared to 8096
pixel output. Here are some sample timings for this
dataset using 16 threads at various resolutions:

16384 pixels
real    18m0.539s
user    29m47.124s
sys     7m15.971s

8096 pixels
real    7m27.724s
user    27m25.347s
sys     0m11.493s

4096 pixels
real    3m19.724s
user    20m34.954s
sys     0m11.546s

2048 pixels
real    3m9.095s
user    24m0.784s
sys     0m11.290s

If more than a single plots uses the same data grid and map projection,
time for subsequent plots can be much less than the first plot.

We are actively working on further improvements and we think that
additional efficiencies are definitely possible. If you have particular
high-res data that you would like us to be able to plot in a reasonable
amount of time, we would like to work with it. If that is the case, please
contact me.

- Dave Brown
NCL development team
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/parvis-users/attachments/20140806/f71e351d/attachment.html>