Hi,
I run a Lagrangian simulation in parallel but the way the domain is split doesn't help, is there some way to tell code_saturne how he should split the domain?
Split a domain as I wish
Forum rules
Please read the forum usage recommendations before posting.
Please read the forum usage recommendations before posting.
Re: Split a domain as I wish
Ok, I found some options in the GUI in Calculation management/Performance tuning, but none gives me what I want.
My domain is a straight duct and I'd like partitions to be long and thin in the X direction. No ideas?
My domain is a straight duct and I'd like partitions to be long and thin in the X direction. No ideas?
-
- Posts: 4208
- Joined: Mon Feb 20, 2012 3:25 pm
Re: Split a domain as I wish
Hello,
No, there is not. What you want would possibly slightly improve performance for the Lagrangian part, but degrade it for the Eulerian parts, which are still a big part of your computation (unless you are running with a frozen velocity field).
The only solution (short of modifying src/mesh/cs_partition.c) would be to use the "block" (unoptimized) partitioning, if you can ensure that the original mesh is numbered in such a way that splitting cells in equal-sized block based on their initial numbers leads to the partitioning you want.
Regards,
Yvan
No, there is not. What you want would possibly slightly improve performance for the Lagrangian part, but degrade it for the Eulerian parts, which are still a big part of your computation (unless you are running with a frozen velocity field).
The only solution (short of modifying src/mesh/cs_partition.c) would be to use the "block" (unoptimized) partitioning, if you can ensure that the original mesh is numbered in such a way that splitting cells in equal-sized block based on their initial numbers leads to the partitioning you want.
Regards,
Yvan
Re: Split a domain as I wish
Thank you for your answer,
I indeed run in frozen velocity field so my Eulerian computation is a separate issue.
The block partitioning seems the best option for my special case, I'll see if I can rework my mesh numbering efficiently.
I indeed run in frozen velocity field so my Eulerian computation is a separate issue.
The block partitioning seems the best option for my special case, I'll see if I can rework my mesh numbering efficiently.