multiprocessing

[bewegende-Architektur]

multiprocessing

[JWock82]

This is an interesting idea. I briefly looked into multiprocessing about a year ago. I think it's possible, but a lot of work. Right now the program is pretty fast for most of the models I do. I could see the utility of multiprocessing for very large models with lots of nodes.

I haven't run a profiler on the code since I added physical members, but the slowest part of the code used to be the assembly of the global stiffness matrix. Multiprocessing that would be difficult, since the total matrix is the sum of the member matrices, and you can only add one member matrix at a time to the global matrix. All the matrix operations have to get in a single file line. I suppose you could have multiple global matrices at a time being created, and then add them together at the end in a few operations. I'm sure there's a way to do it. It seems like we'd lose a lot of the simplicity the code currently has, but maybe that's necessary if we ever want to efficiently solve really big models with tens of thousands of nodes.

I wouldn't be surprised if descretizing physical members lagged on large models, since each node in the model is checked against each physical member in the model to determine if it lands on that member. I think there's a better way to do this that doesn't require revisiting every node on every member, but for now it gets the job done.

There is definitely room for improving these algorithms to be more efficient. Some of them were put together to just get the job done right, without much thought of efficiency. I am a structural engineer, rather than a programmer. Adding multiprocessing is a matter of profiling the code on a large model to find the choke points, and then figuring out how to feed those processes to multiple cores.

[bewegende-Architektur]

Ok. I can answer the question: Members and phys_member would need to be also of type manager.dict()
Guess this is a huge amount of work. I will do another quick test ...

[bewegende-Architektur]

Thanks for your quick reply! Totally agree. I was just wondering, if there is an easy approach.
Greetings,
Chris