Peculiar issue with running REAL.EXE in parallel

Having problems running the WRF Preprocessing Sysmte (WPS)?

Peculiar issue with running REAL.EXE in parallel

Postby megajoule_inov » Mon Oct 15, 2012 1:11 pm

Hello,

A peculiar issue is occurring whilst running REAL.EXE in parallel.
REAL.EXE is producing the exact same boundary fields for the y-start and y-end boundaries (as checked by using ncview to look at wrf_bdy files and for all quantities, say V_BYS=V_BYE, and all time steps), when it is invoked in parallel mode with 7 processes or multiples thereof (14, 28, etc.)

The end result is a WRF simulation at odds in terms of speed and direction with local simultaneous wind measurements.

However, when a different number of processes is selected, say, 1, 3, 6, etc., the problem disappears!
This happens with WRF 2.2 and WRF 3.3.1, on a cluster of 8-core blades, in case it's relevant.

Anyone experienced this or something akin?
Would you think it a bug in REAL.EXE, in the mpi implementation (MPICH 1.2.1, btw), some other issue?

Thanks,
Carlos Silva Santos
Megajoule Inovação
megajoule_inov
 
Posts: 1
Joined: Mon Oct 15, 2012 12:56 pm

Re: Peculiar issue with running REAL.EXE in parallel

Postby jan » Tue Oct 16, 2012 4:48 am

This is a very specific question. I would suggest to give this information to wrfhelp@ucar.edu . Maybe they can tell you something about this behaviour or fix the bug.
jan
 
Posts: 174
Joined: Wed Jun 15, 2011 3:48 am
Location: Kiel, Germany, GEOMAR


Return to Runtime Problems

Who is online

Users browsing this forum: No registered users and 5 guests