[bastil]

Originally Posted by hjasak

Actually, I've got an update for you. There is a new layer of optimisation code built into the GGI interpolation, aimed at sorting out the loss of performance in parallel for a large number of CPUs. In short, each GGI will recognise whether it is located on a single CPU or not and, based on this, it will adjust the communications pattern in parallel.

This has shown good improvement on a bunch of cases I have tried but you need to be careful on the parallel decomposition you choose. There are two further optimisation steps we can do, but they are much more intrusive. I am delaying this until we start doing projects with real multi-stage compressors (lots of GGIs) and until we get the mixing plane code rocking (friends involved here).

Further updates are likely to follow, isn't that right Martin?

Thanks Hrv,

that sounds great. If you need some support with testing this with large models please let me know and I will run it on some of our models.
Is there a rough idea about how and when this will make it into 1.5-dev/1.6-dev? Thanks.

Regards Bastian

[bastil]

Hrv, Martin,

I am just wondering how the ggi speedup work is proceeding? As far as I see there are now news about that in 1.6-ext so far? I started working around ggis to get my cases run in OpenFOAM which works but is a bad compomise in meshing.

Regards Bastian

Originally Posted by bastil

Thanks Hrv,

that sounds great. If you need some support with testing this with large models please let me know and I will run it on some of our models.
Is there a rough idea about how and when this will make it into 1.5-dev/1.6-dev? Thanks.

Regards Bastian

[mbeaudoin]

Hello Bastian,

This work is progressing slowly. On my side, the mixingPlane and some other non-OpenFOAM activities are sucking all my available time.

Some improvements are coming though for speeding up the GGI.

I have developed a replacement to the slow AABB neighbours search algorithm for the GGI facets. The new search algorithm is based on octrees, and is providing a 10x speedup. This will help when initializing the GGI interface, or when using the GGI with moving meshes. I have sent the code to Hrv so he can integrate this with 1.6-ext.

As for speeding up the GGI in parallel, I know Hrv is working on this, and I will too as soon as I am done with the mixingPlane.

The GGI is slow for my research group as well, so this is certainly a high priority for me.

Best,

Martin

Originally Posted by bastil

Hrv, Martin,

I am just wondering how the ggi speedup work is proceeding? As far as I see there are now news about that in 1.6-ext so far? I started working around ggis to get my cases run in OpenFOAM which works but is a bad compomise in meshing.

Regards Bastian

[bastil]

Originally Posted by mbeaudoin

Some improvements are coming though for speeding up the GGI.

I have developed a replacement to the slow AABB neighbours search algorithm for the GGI facets. The new search algorithm is based on octrees, and is providing a 10x speedup. This will help when initializing the GGI interface, or when using the GGI with moving meshes. I have sent the code to Hrv so he can integrate this with 1.6-ext.

Martin, Hrv,

I see some changes made their way into 1.6-ext. I am currently compiling it. I see that the MixerGgi tutorial still uses globalFaceZones and therefore I think that this is still necessary? Which of the changes are already in and wich are still to come? Thanks.

Bastian

[arjun]

Originally Posted by mbeaudoin

The GGI is slow for my research group as well, so this is certainly a high priority for me.

how slow is your algorithm at the moment. Lets say for 100K polygonals how much time you are taking?? (in seconds).

Just wondering.

[Ohbuchi]

Hi,

I've found the new update for GGI parallel performance improvement on the Git hub site.
I was really looking forward to this, so I've download these update, and recompile the OpenFOAM-1.6-ext and tested my test case with GGI interface (see attatched image).

But, in my test case (in parallel), computational time was much longer than the first release of OF-1.6-ext.
I was really disappointed. Or, Did I made some mistake for setting new feature in GGI?

[bastil]

Originally Posted by Ohbuchi

But, in my test case (in parallel), computational time was much longer than the first release of OF-1.6-ext.
I was really disappointed. Or, Did I made some mistake for setting new feature in GGI?

How did you set it up? Did you use GloblaFaceZones for your ggis when decomposing? I am not really sure if this is still essential.

Regards Bastian