GrADS bad speed performances

Arlindo da Silva arlindo.dasilva at GMAIL.COM
Tue Sep 22 10:13:00 EDT 2009


On Wed, Sep 2, 2009 at 1:51 PM, Ivan Toman <ivtoman at inet.hr> wrote:

> Hello. I have very strange issue and help is greatly appreciated.
>
> I run WRF-NMM 36-12-4km nesting and after model finishes I forward
> output grib1 files to grads (1.9) from all three domains in successive
> order. First, 36km domain is processed in GrADS, when it finish 12km
> domain comes to work and when it finish, 4km go into GrADS at the end.
>
> My problem is that GrADS has good performance only on 12km domain, on
> other domains it is very and extremely slow. These are times for
> creating map graphics from these domains:
>
> 36km - 40 seconds
> 12km - 6 seconds
> 4km - 105 seconds
>
> It is very clear something very strange is going on. Why last domain
> gribs files needs almost two minutes to postprocess in GrADS, when
> intermediate domain gribs takes only 6 seconds to complete the same job?
>

It has been my experience that you need to run these jobs at least 20 times
each and average the execution times in order to get stable statistics.
Also, process more than 1 time step at a time to factor out the time it
takes to start grads. An execution time of 6s is too short not to get
entagled with the usual latency of starting grads.

          Arlindo




> Yes, for all three domains is executed completely same .gs script,
> domain sizes are almost the same (in number of points), and GrADS don't
> output any error or similar unusual messages on any domain. There are no
> other CPU jobs that could slow down processing. GrADS use one CPU core
> at 100% all of time during this hanging.
>
> These times are for only 6 output gribs (still testing), I need to
> extend forecast now to 60 hours, so there will be 10 times more data
> than now is, if now GrADS process for 105 seconds then it will work 17.5
> minutes to draw all needed maps from 4km domain?? This will become a
> serious concern then.
>
> Please, can anybody help me to resolve this problem?
>
> Thanks
>



--
Arlindo da Silva
dasilva at alum.mit.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://gradsusr.org/pipermail/gradsusr/attachments/20090922/fa63f826/attachment.html 


More information about the gradsusr mailing list