meshing on a cluster
Hey everybody!
Does anybody know how meshing does work on a cluster? Calculating is no problem, but just meshing?!:( Thank you for your help! Jopawipr |
Meshing is still serial, and not parallel. I haven't tried it yet, but I assume it is done on the controller if you're running in parallel.
|
Some meshing functions can run parallel but not all :(
|
Meshing on a cluster
Hello,
I do this fairly often. It works well if you have a node that has more memory then the rest, which is the case on our clusters. I typically launch a job that only does the meshing and saves the file then I run the job later. I request a node by using a memory request in the job submission script then the queue manager directs the job to the node with more memory. That way if the job crashes half way through you don't have to remesh. I have run jobs the remesh based on the flow field solution gradients during the solution but this is time consuming and not appropriate for all types of jobs. As others have said meshing is still primarily serial, which is a major weakness of the mesher IMHO. Todd |
Split the regions and mesh per region, then you can mesh with less memory.
|
All times are GMT -4. The time now is 06:33. |