basics of MPI

Our local cluster configuration consists of two Rocketcalc clusters, managed by a dual processor workstation. The cluster consists of this workstation (idefix) and 8 nodes. Every node has 1Gb internal memory. As four nodes are dual processors, the cluster has 14 processors, every one running at a clockspeed of 2.4Ghz.

We described the various message-passing programming options and defined the distinction processes and processors, outlining the SPMD programming model.

Our first program is of course Hello World!. A screenshot of the session goes as follows: 

prompt] mpicc hello.c -o hello
prompt] mpirun -np 10 hello
Hello world from processor 0 out of 10.
Hello world from processor 2 out of 10.
Hello world from processor 4 out of 10.
Hello world from processor 8 out of 10.
Hello world from processor 6 out of 10.
Hello world from processor 1 out of 10.
Hello world from processor 7 out of 10.
Hello world from processor 5 out of 10.
Hello world from processor 3 out of 10.
Hello world from processor 9 out of 10.
prompt]
We illustrated the broadcast with broadcast_integer.c and broadcast_doubles.c. Note that, as we are working in a distributed memory computer, every node (not just the manager node) must allocate memory for the array of doubles.

Bibliography