Escali Escali, LLC Network Router 4.4 Manual Do Utilizador

Section: 3.3 Running Scali MPI Connect programs

<pid> 

is the Unix process identifier of the monitor program mpimon.

<nodename> 

is the name of the node where mpimon is running.

Note: SMC requires a homogenous file system image, i.e. a file system providing the same 
path and program names on all nodes of the cluster on which SMC is installed.

The control and start-up of an Scali MPI Connect application are monitored by mpimon. A 
complete listing of mpimon options can be found in “Mpimon options” on page 34.

Normally mpimon is invoked as:

mpimon <userprogram> <program options> -- <node name> [<count>][<nodename> 
[<count>]]...

where

is name of application

<program options> are options to the application
-- 

is the separator ending the application options

<nodename>[<count>] is name of node and the number of MPI-processes to run on 

that node. 

The option can occur several times in the list. Mpi-processes will 

be given ranks sequentially according to the list of node-number 
pairs. 

The <count> is optional and defaults to 1

Examples:
Starting the program “/opt/scali/examples/bin/hello” on a node called “hugin”:
mpimon /opt/scali/examples/bin/hello -- hugin

Starting the same program with two processes on the same node:
mpimon /opt/scali/examples/bin/hello -- hugin 2

Starting the same program on two different nodes, “hugin” and “munin”:
mpimon /opt/scali/examples/bin/hello -- hugin munin

Starting the same program on two different nodes with 4 processes on each:
mpimon /opt/scali/examples/bin/hello -- hugin 4 munin 4

Bracket expansion and grouping (if configured) can also be used :
mpimon /opt/scali/examples/bin/hello -- node[1-16] 2 node[17-32] 1

for more information regarding bracket expansion and grouping, refer to Appendix D.

The identification of nodes and the number of processes to run on each particular node 
translates directly into the rank of the MPI processes. For example, specifying n1 2 n2 2 will 

place process 0 and 1 on node n1 and process 2 and 3 on node n2. On the other hand, 
specifying n1 1 n2 1 n1 1 n2 1 will place process 0 and 2 on node n1 while process 1 and 3 

are placed on node n2.