[Wrf-users] Single node is faster that 2 nodes on IBM.

[mti18 at student.canterbury.ac.nz]

Dear WRF users,
 
I have an original question: I use ARW WRF compiled with option 4 for IBM
computer (OBJECT_MODE = 64).
I ran the same test case  on 1 node (16 processors) and on 2 nodes (32
processors). The results were stunning for me: 1 hour on 1 node (16 prosc)
runs 2.30 minutes and 1 hour on 2 nodes (32 procs) runs 10 minutes!!! So far
for ARW WRF (option 4, IBM) in my case 1 node 4 times faster than 2 nodes!?
My system operator tries to argue that this is a fault of ARW WRF
model:something like 'that a split of a quite small domain (XMAX = YMAX =
34, 4 grids) for 2 nodes destroys all advantages of using 32 processors and
makes the things even worse (brakes a speed of WRF run).
 
Does anybody have any experience about such kind of 'reverse' events?
Is the guess of my system administrator is right or is there something wrong
with IBM initial configuration? 
 
Any help is very highly appreciable.
Many thanks,



^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Mikhail Titov
Doctoral Candidate
Department of Geography
University of Canterbury - Te Whare Wananga o Waitaha
Private Bag 4800
Christchurch
New Zealand
Telephone:  +64 3 364 2987 ext 8839
Fax (Department): +64 3 364 2907
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 

 

 
<http://promos.hotbar.com/promos/promodll.dll?RunPromo&El=&SG=&RAND=76388&pa
rtner=hbtools> Upgrade Your Email - Click here! 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/wrf-users/attachments/20061103/ad9f80f3/attachment.html