WIN GRADS 2.0.a7.oga.3 - Possible Bug

[Andrew Revering]

Yeah that's just the way windows reports it. In reality, the process is in
one core, which maximizes the 'processor' at 50% if it’s a dual core
machine...even though the graphs show it on two cores.

With the help of Arlindo I have a very neat setup in place to create a
national radar composite. I had to be creative to make it timely using an 8
thread machine (4 cores * 2 (hyperthreading))... I broke up the script into
8 sub-scripts. Each launched from a proprietary app calling the GrADS
scripts from a shell command. Windows opens each instance of GrADS in its
own core, so you get to utilize all cores this way.

I increased my processing time on an I7 920 processor from roughly 28
minutes down to 5 minutes, and was able to increase the resolution in the
process as well.

Andy

-----Original Message-----
From: GRADSUSR at LIST.CINECA.IT [mailto:GRADSUSR at LIST.CINECA.IT] On Behalf Of
Ryo Furue
Sent: Monday, January 25, 2010 9:22 PM
To: GRADSUSR at LIST.CINECA.IT
Subject: Re: WIN GRADS 2.0.a7.oga.3 - Possible Bug

Hi Mark,

| On another subject....Multi threading.  When I run Grads 1.9 on a
| dual core machine, it appears to split the work between the two
| cores with each core getting about 50-54% of the load.  On a single
| core machine it takes 100% of available CPU.   So it splits the
| load, but does not make 100% use of both CPUs. 

Since I'm not sure how Windows gather statistics like that,
this is a pure guess, but I guess GrADS is frequently switched
between the two cores; that is, it is run sometimes on
one core and sometimes on the other.  (GrADS is not
multi-threaded, I think.)  So, if you take an average,
you will see each 50% of each core is used.

Linux kernel doesn't migrate a process from one core
to another that often, so that you actually see that
one process uses 100% of one core at one moment and
the same process uses 100% of another core at another
moment.

Regards,
Ryo