問題描述
在 ARM 上的 Linux 上運行 MPI (OpenMPI) 應用程序時出現問題 (Problems running MPI (OpenMPI) app on Linux on ARM)
I am trying to follow this tutorial for building and running an MPI application on an ARM based Ubuntu 11.10 system.
When installing open‑mpi environment on my PC machine, the sample program runs well. However, trying the same on the ARM machine, the terminal hangs up and I need to kill the MPI process from a second terminal in order to release it.
The MPI packages I installed using apt‑get, on both machines, were mpi‑default‑dev
and mpi‑default‑bin
, so I assume that the packages are as updated as they can be.
The first sample program in the tutorial makes every process prints a "hello" message with some info. On the PC I get messages from all 8 processes (although running on a single core) and then the program ends. On the ARM, I get no output at all. The program is just stuck immediately after launch.
Any idea on what's wrong? I am not sure even where to start to debug this?
Update: I tried removing the OpenMPI package and install the alternative MPICH2 package ‑ but the result is just the same.
‑‑‑‑‑
參考解法
方法 1:
Ubuntu 11.10 did not ship with a functional Open MPI implementation for ARM (although it may have shipped with a nonfunctional one). Ubuntu 12.04 did.
I would recommend building your own Open MPI from source ‑ available at http://www.open‑mpi.org/software/ompi/v1.6/, unless you can update to a more recent version of Ubuntu.
Alternatively, you could rebuild the 11.10 package using the fixes pointed out in https://bugs.launchpad.net/ubuntu/+source/openmpi/+bug/949044.