One quickly appreciates the limits of a machine while doing urban scale outdoor CFD studies. The machine I am using has 32 GB ram, i7 3.6 GHz 4 cores. Even after using a clever meshing technique introduced by @TheodorosGalanos called “Gradient mesh”, there are instances on a regular basis that I have to omit geometry to reach a mesh level with which I can run solution in an acceptable time.
I am aware that @TheodorosGalanos knows a way to run solution over the cloud. I don’t think I can take that route. So is there any other way to take advantage of a typical network of computers at workplace?
I expect the process will be complicated at the least. So if someone can point me to a resource where I can learn to do this I would really appreciate that.
In one sentence, yes you can use a network of computers running linux (Beowulf) passing tasks with mpirun using a machine file. These keywords should get you a few articles on a google search.
However, unless you have a fiber optic network I wouldn’t suggest that. You will be losing a lot of time in communication through the network, even with gigabit networks. I think I would first try and fit things in memory. I regularly use a similar set up (memory is the only real bottleneck) for quite demanding studies and I usually find ways to fit it. It’s not always ideal but with some trial and error you should be able to find some good compromise.
If you do think you’ll be running many CFD studies in the future, I’d invest a bit of money in 32gb more, when you can.