Hello,
I'm trying to run also demo.r as described in this thread and encountered the same problem John Alexander described.
I checked pbdMPI vignettes, Section 8.1, FAQ, question 8 as suggested and found there might be problems when having MPICH and openMPI both installed, so I removed MPICH from my cluster. Now I only have openmpi:
$ apt-show-versions
...
libopenmpi-dev:amd64/trusty 1.6.5-8 uptodate
libopenmpi1.6:amd64/trusty 1.6.5-8 uptodate
openmpi-bin:amd64/trusty 1.6.5-8 uptodate
openmpi-common:all/trusty 1.6.5-8 uptodate
However, I can't run the demo.r example either. As John Alexander described, when I run the program it does not finish but the CPU is very busy executing a R process.
To test the cause is a communication problem, I tried another example: 'hello.r' with the following code
print("hello")
print(Sys.info()["nodename"])
I get a normal result when running it
$ mpiexec --hostfile my_hostfile -np 2 Rscript hello.r
However, if hello.r is extended to include some pbdR code as follows:
print(Sys.info()["nodename"])
comm.print(x, rank.print=0)
When running it, I get
$ mpiexec --hostfile my_hostfile -np 2 Rscript hello.r
[1] "hello"
nodename
"master"
[1] "hello"
nodename
"slave"
and it never finishes... So the communication is working but apparently there is an issue with pbdMPI which I am not able to solve.
Could anybody kindly help me?
Thanks,
Cristina