CFD Online Discussion Forums - decomposePar over several disks

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)

- OpenFOAM (https://www.cfd-online.com/Forums/openfoam/)

- - decomposePar over several disks (https://www.cfd-online.com/Forums/openfoam/93498-decomposepar-over-several-disks.html)

decomposePar over several disks

Greeting Foam Users,

I'm working on parallelization some of our cases and solvers and was looking to ask questions about distributing data over several disks. I have two multi-core machines I have made into "cluster" and am testing decomposed cases. Just using the built in network cards to communicate (these are Dell Precision T5500 with Broadcom BCM5761 and an ethernet crossover cable). what are some tips to reduce the overhead so that communication quick, or some inexpensive modifications that I could make to improve performance.

I was also curious about different methods for domain decomposition. Specifically, how the manual option works (with examples if possible), and if the simple/hierarchical methods can be controlled by subdomain (ie, after the first division of a domain in two equal parts, can I further subdivide one of the sub-domains)?

Thanks!

Hi Marco,

I'm going to have to be quick... so here goes...

The documented version of said dictionary can be found this way:

Code:

find $WM_PROJECT_DIR/applications -name decomposeParDict

Here you'll find a lot about this dictionary :D

All other examples can be found this way:

Code:

find $WM_PROJECT_DIR/tutorials -name decomposeParDict

As for dumping the cell<->processor association, I believe this should help: http://openfoamwiki.net/index.php/Ti...gisteredObject

As for a whole lot of other notes about OpenFOAM+parallel: http://www.cfd-online.com/Forums/blo...-parallel.html

Best regards,
Bruno

There was a message here about ParaView having a "decomposed case" option, but that has since been deleted. I looked at my current version of ParaView (3.10.1) and was not able to finde the decomposed case option. Does anyone else know about this?

With the paraFoam -touch option, I am able to load all the processors into one instance of ParaView and do post-processing on the entire domain. The only problem is that application of filters has to be done on each processor in turn. Fine for domains of 5 processors or less, but this could get unwieldy (at least the way I'm doing it). We can apply multiple filters to single sources, can we apply a single filter and take as input multiple sources?

Hi Marco,

Quote:

Originally Posted by mturcios777 (Post 328812)

Strange... why would it be deleted? Wasn't it moved?
But if you are using OpenFOAM 2.0, run paraFoam like this:

Code:

paraFoam -builtin

It will open the case with the internal reader that ParaView has, namely using the file extension ".foam", instead of ".OpenFOAM".

Quote:

Originally Posted by mturcios777 (Post 328812)

With the paraFoam -touch option, I am able to load all the processors into one instance of ParaView and do post-processing on the entire domain. The only problem is that application of filters has to be done on each processor in turn. Fine for domains of 5 processors or less, but this could get unwieldy (at least the way I'm doing it). We can apply multiple filters to single sources, can we apply a single filter and take as input multiple sources?

There's a filter that will join multiple sources into a single source... I think it's "Group Datasets". Then you can simply manipulate the resulting object.

Best regards,
Bruno

Beautiful. I need to play with that feature a bit more, as it seems to have some trouble synchronizing the time directories (it only gives me 0, 0.1111... and 1). The group datasets feature does what I need, albeit with a little bit more work; I'll use it until I can get the built-in reader to work properly.

Much thanks!