transient, parallel - problem with timestep
 

hello,

i've a model with transient rotor stator interface, running in transient parallel mode.

problem is, that if i do it with two cpus everything is fine and i need 5 inner loops for each time step, but if i use more 8 cpus the inner loops don't converge anymore?!

has anyone observed a similar behavior?

has anyone an idea what the problem is?

are there any restrictions in parallel, e.g. that the whole interface has to be on one cpu or so?

kind regards

peter

Re: transient, parallel - problem with timestep
 

Hi,

Are you using the current version of CFX (5.7.1)? Some older versions of CFX had problems in multi-processor mode at GGI's for transient rotor-stator models. The current version should be OK.

Glenn Horrocks

Re: transient, parallel - problem with timestep
 

hello glenn,

the cfx version is 5.7.1 Build 2004.11.12-23.20

regards

peter

Re: transient, parallel - problem with timestep
 

Hi Peter,

Can you describe your model a bit more? Are you modelling a free surface?

Glenn Horrocks

Re: transient, parallel - problem with timestep
 

hello glenn,

the model is a turbine with a transien rotor stator interface between guide vane and runner. approx. 4.000.000 hexaeder cells. 4 domains.

made some tests in between and found out that if i take a different method for decomposition then it works even with more cpus, but the partition is bad now (weigt 0.3 for the first cpu and weight 0.1 for all the others) - i used methis again, but changed the multi domain option from the default to 'coupled partitioning'

regards

peter

Re: transient, parallel - problem with timestep
 

Hi Peter,

Your problem sounds a bit strange. I would contact your support rep, you may have found a bug. It would need a detailed look at your run to establish exactly what's happening and you can't really do that on this forum.

Glenn Horrocks