Dear Francis, On 08/02/16 20:48, Francis Poulin wrote:
I have downloaded some demos and am starting to learn how they work. I learned that to run it in parallel I need to specify the path for my firedrake version of python
/usr/local/bin/mpirun -np 4 ~/software/firedrake/bin/python burgers.py
These run in parallel and do seem to take up the right number of cores. I am curious to do some tests on my laptop, desktop, and cluster to see how things scale. The actual problem doesn't matter so much but I would like it to be something like Shallow Water, Navier Stokes, or something in that line.
I see you have found the firedrake-bench codes which provide some of this. We used some of these codes when running the scaling benchmarks in the firedrake paper (http://arxiv.org/abs/1501.01809). Section 6 of that paper characterises the performance of firedrake on a variety of problems, and in particular explores the strong and weak scaling limits. Thanks, Lawrence