|
[Sponsors] |
June 13, 2007, 21:59 |
Multi domain performance penalty????
|
#1 |
Guest
Posts: n/a
|
Can someone tell me if there is a performance penalty when using multiple domains in a sim. Example, I have a large tank, that I have meshed using icem and cfx-mesh. CFX-mesh for the more complex inlet pipes and Icem for the bulk of the rest of the tank. I join the domains together in cfx-pre and all works well.
Neglecting any difference in nodes/elements/tets/hexes, would the solution take longer to solve because it is using 2 doamins, or would it be faster if I did the entire mesh using 1 domain. Thanks Stu |
|
June 14, 2007, 06:43 |
Re: Multi domain performance penalty????
|
#2 |
Guest
Posts: n/a
|
In the problems I've solved I've noticed little to no penalty. These were simple linear geometries with a domain interface in the middle.
|
|
June 14, 2007, 19:41 |
Re: Multi domain performance penalty????
|
#3 |
Guest
Posts: n/a
|
The GGI's will deliver a small penalty.
For parallel processing the penalty can be larger. CFX has the drawback that each partition needs at least a part of each domain. So when you have 10 small parts and 1 large part and 3 processors, these 10 small parts are partitioned over the 3 processors as well as the large parts. Even if the 10 small parts would fit on 1 processor. Gert-Jan |
|
June 14, 2007, 19:46 |
Re: Multi domain performance penalty????
|
#4 |
Guest
Posts: n/a
|
Hi,
Whether multiple domains decreases efficiency is problem dependant. The issues are: Increased efficiency due to: 1) Possibility of improved mesh quality and sizing compared to single domain Decreased efficiency due to: 1) In parallel, partitioning is done on a domain by domain basis (by default) so there the partition interface area increases. 2) There is an overhead in interfaces. 1 to 1 interfaces have almost no overhead but GGI interfaces have a small overhead. The balance of these issues determines the effect of the interfaces. Glenn Horrocks |
|
June 14, 2007, 23:36 |
Re: Multi domain performance penalty????
|
#5 |
Guest
Posts: n/a
|
By default, CFX runs in "Independent Partitioning" which behaves as Gert-Jan describes. However, you can run "Coupled Partitioning" and then the partitions do not have this restriction.
|
|
June 16, 2007, 14:30 |
Re: Multi domain performance penalty????
|
#6 |
Guest
Posts: n/a
|
Hmmmm, learning every day. Since when does this option exists?
Gert-Jan |
|
June 18, 2007, 06:46 |
Re: Multi domain performance penalty????
|
#7 |
Guest
Posts: n/a
|
I think it has been an option for a long time. Definitely it was there in 10.0.
|
|
June 18, 2007, 08:32 |
Re: Multi domain performance penalty????
|
#8 |
Guest
Posts: n/a
|
I started with CFX 5.3. It was not present there. Hard to keep up all the new features.......
Gert-Jan |
|
June 18, 2007, 19:57 |
Re: Multi domain performance penalty????
|
#9 |
Guest
Posts: n/a
|
Hi,
Yes, it has been there for a while. I have played with it a long time ago for my simulations with a few GGIs in the hope it increased the simulation speed but it actually made it marginally worse. Not sure why. I would be interested to hear if it works better now, or if it is an improvement for some simulations and not for others. Glenn Horrocks |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Domain Imbalance | HMR | CFX | 5 | October 10, 2016 06:57 |
CFX domain comparison | Kiat110616 | CFX | 4 | April 3, 2011 23:43 |
Error Message after switching to multi domain physics | chili023 | CFX | 3 | June 5, 2010 06:28 |
Multi flow in two different domain | arunss | CFX | 3 | May 11, 2010 17:06 |
multi domain heat transfer prob | mohammed zubair | CFX | 1 | June 27, 2006 10:30 |