ParMETIS / intel MPI run time issue since 7.2

gerald · February 7, 2022, 03:38

Hello,
I face trouble having a working parallel installation of SU2 version 7.2 and 7.3 with intel MPI (19 or 21). The problem shows up at run time from the ParMetis with execution errors:

PARMETIS ERROR adjncy is NULL.
PARMETIS ERROR adjncy is NULL.
PARMETIS ERROR adjncy is NULL.
187966 boundary elements in index 3 (Marker = wing).
Executing the partitioning functions.
Building the graph adjacency structure.
Calling ParMETIS...

We did not encountered such a problem in previous build (7.0 and even 7.1), with the same (INTEL) MPI libraries.
Any idea of where this issue could come from ? Any change in ParMetis coming with SU2 between 7.1 and 7.2 ?
Thanks in advance for your help!
Gérald

Cherith · January 29, 2023, 15:37

I am getting the same error for higher order triangular elements, as the node ordering is different the way gmsh does and SU2 interprets, so i changed the order of nodes for connectivity but now i am getting a segmentation fault donno why, could anyone suggest a way to solve this...

bigfootedrockmidget · January 30, 2023, 04:38

so does the exact same setup work for you for lower order elements? Can you make the 2 cases available,

1. working setup with linear elements
2. same setup with nonlinear elements crashing

I personally do not know too much about this implementation, but it will definitely help to have a practical problem to look at and to be able to reproduce it.

Cherith · July 16, 2023, 13:15

So the issue was that lets say there is a second order triangular element, when we solve a problem in FEM the number is like
3
65
142 right , but when we output a mesh from gmsh it numbers it as

6
45
123 . And it messes up the type of element it mentions in the file, instead of giving out the element type to be 109 which is the reference for a 2nd order square element it gives out 28 and even for the boundary markers it gives out something else.
So these things when SU2 tries to read the mesh becomes a problem and we get the connectivity problems.

It would be a lot easier if someone could fix the code in gmsh to give out the right mesh configuartions for higher order meshes.

February 7, 2022, 03:38	ParMETIS / intel MPI run time issue since 7.2	#1
gerald New Member Join Date: Apr 2021 Posts: 6 Rep Power: 5	Hello, I face trouble having a working parallel installation of SU2 version 7.2 and 7.3 with intel MPI (19 or 21). The problem shows up at run time from the ParMetis with execution errors: PARMETIS ERROR adjncy is NULL. PARMETIS ERROR adjncy is NULL. PARMETIS ERROR adjncy is NULL. 187966 boundary elements in index 3 (Marker = wing). Executing the partitioning functions. Building the graph adjacency structure. Calling ParMETIS... We did not encountered such a problem in previous build (7.0 and even 7.1), with the same (INTEL) MPI libraries. Any idea of where this issue could come from ? Any change in ParMetis coming with SU2 between 7.1 and 7.2 ? Thanks in advance for your help! Gérald

January 29, 2023, 15:37	Problem with higher order mesh in SU2	#2
Cherith New Member Cherith Lavisetty Join Date: Jan 2023 Posts: 7 Rep Power: 3	I am getting the same error for higher order triangular elements, as the node ordering is different the way gmsh does and SU2 interprets, so i changed the order of nodes for connectivity but now i am getting a segmentation fault donno why, could anyone suggest a way to solve this...

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Postprocess: sampleDict works but creates no output folder	shock77	OpenFOAM Post-Processing	14	November 15, 2021 08:27
[blockMesh] Issue with mergePatchPairs	JohnnyMercu	OpenFOAM Meshing & Mesh Conversion	1	May 30, 2021 00:24
laplacianFoam with source term	Herwig	OpenFOAM Running, Solving & CFD	17	November 19, 2019 13:47
pimpleDyMFoam computation randomly stops	babapeti	OpenFOAM Running, Solving & CFD	5	January 24, 2018 05:28
Micro Scale Pore, icoFoam	gooya_kabir	OpenFOAM Running, Solving & CFD	2	November 2, 2013 13:58

January 30, 2023, 04:38		#3
bigfootedrockmidget Senior Member bigfoot Join Date: Dec 2011 Location: Netherlands Posts: 505 Rep Power: 17	so does the exact same setup work for you for lower order elements? Can you make the 2 cases available, 1. working setup with linear elements 2. same setup with nonlinear elements crashing I personally do not know too much about this implementation, but it will definitely help to have a practical problem to look at and to be able to reproduce it.

July 16, 2023, 13:15		#4
Cherith New Member Cherith Lavisetty Join Date: Jan 2023 Posts: 7 Rep Power: 3	So the issue was that lets say there is a second order triangular element, when we solve a problem in FEM the number is like 3 65 142 right , but when we output a mesh from gmsh it numbers it as 6 45 123 . And it messes up the type of element it mentions in the file, instead of giving out the element type to be 109 which is the reference for a 2nd order square element it gives out 28 and even for the boundary markers it gives out something else. So these things when SU2 tries to read the mesh becomes a problem and we get the connectivity problems. It would be a lot easier if someone could fix the code in gmsh to give out the right mesh configuartions for higher order meshes.