Hi, dear bempp team, I am very appreciated about this excellent code and the platform which we can communicate with each other about BEM.
I am planning to verify the capability of Bempp in simulating large-scale acoustic scattering problems. I have tested 13,0000-DOF case currently. It works very well on a single node of my cluster(only CPU Cores) even if the FMM was not activated.
But the RAM on single CPU node seems not enough for several millions DOFs simulation. I am considering using cross-node parallelism, but there has not had any instructions in Bempp yet. I noticed that the MPI support was enhanced in Sep. 2020 (Issue 50 on Bempp-Github), could you please tell me how to use it in Bempp?
Bempp does not yet support MPI. It is in development. But there is no timeline for it. Alternatively, you could use our Exafmm bindings. For low-frequency problems they work quite well. For Laplace we have run simulations of up to 10 million dofs on a single node using FMM.
Hi, thanks so much for your suggestion and shared experience. It is so exciting to hear that the Laplace-type problems with 10 millions dofs can be simulated on a single node using FMM. I am interested in Helmholtz equation for acoustic scattering problems. My colleagues and I want to do some large-scale extensions based on Bempp and mixed-precision arithmetic algorithm for industrial application like airliner noise. The MPI is necessary for us, so if not complicated for us, we want to develop the MPI support for Bempp. But it definitely will be a huge work. Hence, could you please give me some suggestion about the MPI extension based on your experience for Bempp?
Hi, I forgot another issue. It is that the ExaFMM(GitHub - exafmm/exafmm) has not supported ARM V8 yet because the ExaFMM is optimized for X86 AVX. This information is possibly useful for the ARM users like me.