The NATIVE Driver

The NATIVE driver, which is provided to AMEHPCStudyStart script the through the parameter --driver NATIVE does not use a scheduler internally. In this mode, launching the jobs on all the compute nodes is the responsibility of the HPC framework itself, that is, without any queuing system. You must therefore provide two additional parameters:

• The machinefile containing the information about node names and the number of jobs to launch in parallel on each node (--machinefile parameter).

• The remote shell used to launch jobs from one node to another across the cluster (--remote-shell parameter).

These two additional parameters are discussed in more detail in the following sections.

Remote shells

A remote shell is used to launch jobs across compute nodes and can be provided using the --remote-shell parameter. By default, the remote-shell is the scheduler one, it is therefore a mandatory parameter when using the NATIVE driver (no scheduler). There are three possible values for the remote-shell parameter:

• sched: Use the scheduler to launch jobs across nodes. Note This cannot be used with the NATIVE driver as there is no scheduler in this case.

• ssh: Only available on Linux.

• mpi: Use mpiexec to launch jobs across nodes, only available on Windows.

Machinefiles

A machinefile must be given as a parameter of the AMEHPCStudyStart script when using the NATIVE driver. In such cases, the information about compute nodes and available number of CPUs normally given by the scheduler must be provided. The machinefile must contain information about:

• Compute node names. These should be reachable via the network

• Number of parallel jobs for each node (optional)

• Base TCP port to which the STDSIMManager tries to bind on each node (optional)

If the number of jobs is not provided in the machinefile, the jobspernode parameter of AMEHPCStudyStart must be provided.

The format of machinefiles is presented in the following section. Each line should contain the compute node name, the STDSIMManager base TCP port and the number of jobs:

node1:40001 3

node2:40001 3

The machinefile above defines two compute nodes: “node1” and “node2”, with base TCP STDSIMManager port 40001, and 3 CPUs for each node to use during the study.