A host file

The scheduler generates the mayampi.hosts file at runtime before running the jobs in parallel. This file includes a list of cluster nodes that will execute the jobs.

The following is an example of the host file format, which TMG thermal-flow solversexpect, to execute a simulation in parallel on six processes of two nodes. Each node will run three processes:

tmgnode1
tmgnode1
tmgnode1
tmgnode2
tmgnode2
tmgnode2

Example of a host file for the SGE scheduler

Each scheduler controls the number and names of the nodes that will run the simulation. For example, the SGE scheduler stores this information in a file using the parallel environment variable pe_hostfile. The concept of a parallel environment (pe) is specific to the SGE scheduler. In this example, the parallel environment under which the job are submitted is defined as monitor. The SGE scheduler lets you define a start procedure, which occurs the moment before launching the job in the queue. In the parallel environment, the start procedure is defined as:

/shared_nfs/scripts/create_hostfile.sh $pe_hostfile

Next, the following script creates the mayampi.hosts file in the run directory:

#!/bin/bash
#
#  * $1 is the SGE host file name
#
awk '{for (i=0;i<$2;i++) print $1}' $1 > $SGE_O_WORKDIR/mayampi.hosts