PETSc error

[chenyi-maker]

Hello everyone
I run the Firedrake code in parallel on the server. My server configuration is included in Attach File, it has enough for more than 15 cores, but I can only call up to 15 cores for parallelism. If I call more than 15 cores, he will always report a PETSc error: Caught signal number 7 BUS: Bus error , possibly illegal memory access. My Firedrake is integrated using the Docker environment.

Thank you for your answers

[danshapero]

I think what you're running into relates to the what communication mechanism, or device, MPICH gets built with. The default is to build MPICH using the "nemesis" device. Nemesis enables certain shared-memory optimizations, with the idea that it can work much faster on a workstation or a single machine with very little tweaking on the user's part. But the nemesis device has a limitation that you can't oversubscribe it -- you can use only as many MPI processes as you have cores.

You can change how MPICH gets built so that it uses a different communication device when you build Firdrake; see the Firedrake documentation. There's some good information about the different communication devices in section 4 of the MPICH documentation.

[chenyi-maker]

Hello, thank you very much for your answer. What I want to consult is that my server device now has 28*4=112 cores, but I have only invoked 15 cores, which is within the limit. Why does PETSC error still occur?

[danshapero]

Sorry, I should have noticed that! I take it back, I have no idea what's going wrong then.

[chenyi-maker]

It doesn't matter I still appreciate your advice on how to solve the problem.