Appendix: Sample Commands

發行項
07/21/2009

Applies To: Windows HPC Server 2008

The following tables provide sample commands that can be used to submit a variety of jobs, and they explain how these jobs are run by the HPC Job Scheduler Service.

Sample commands for simple and parametric jobs

Command	Result
`Job submit myapp.exe`	Runs `myapp.exe` on a single processor. The undirected output is available in the Output property of the job task.
`Job submit /workdir:\\headnode\MyFolder /stdout:data.out /stdin:data.in myapp.exe`	Runs `myapp.exe` on a single processor by using a head node file share as the working directory. The input and output are redirected to and from files in that directory.
`Job submit /numcores:4-8 /parametric:100 /stdin:input*.dat myapp.exe`	Runs between four and eight simultaneous instances of the `myapp.exe` command. `myapp.exe` is executed 100 times on files input1.dat, input2.dat, input3.dat, through input4.dat.
`Job submit /exclusive:true /requestednodes:MyNode01, MyNode02 myapp.exe`	Runs `myapp.exe` on either `MyNode01` or `MyNode02`. No other command can execute on that node when this command is running.
`Job submit /NodeGroups:TestGroup myapp.exe`	Runs `myapp.exe` on a node in the node group named `TestGroup`.

Sample commands for MPI jobs

Command	Result
`job submit /numcores:6 /stdout:myOutput.txt mpiexec myapp.exe`	Runs `myapp.exe` on six processors with the standard output directed to the file `myOutput.txt` in the user’s home directory on the head node (as defined in the `%USERPROFILE%` environment variable on the head node).
`job submit /numnodes:2 mpiexec myapp.exe`	Runs one `myapp.exe` process on two compute nodes. The standard output is stored in the HPC Job Scheduler Service database in the Output field for the task.
`job submit /numnodes:2 /requestednodes:Node1,Node2 mpiexec myapp.exe`	Runs one `myapp.exe` process on two compute nodes: `Node1` and `Node2`.
`job submit /numnodes:16 mpiexec -cores 4 myapp.exe`	Runs four `myapp.exe` processes on each of the 16 nodes assigned to the job regardless of the number of physical cores on each node.
`job submit /numcores:24 /workdir:\\headnode\MyFolder mpiexec myapp.exe`	Specifies a single, shared working directory for all 24 `myapp.exe` processes in the job.
`job submit /numcores:128 mpiexec -env MPICH_NETMASK 157.59.0.0/255.255.0.0 myapp.exe`	Runs 128 `myapp.exe` processes where the Message Passing Interface (MPI) data traffic is routed to a specific network in the cluster (157.59.x.x/255.255.0.0 in this example) by setting an environment variable (`MPICH_NETMASK`) on the MPI processes.
`job submit /numcores:16 mpiexec –affinity myapp.exe`	The Windows operating system optimizes the use of multicore systems by dynamically shifting workload to underutilized cores. However, this refinement can be detrimental to the performance of some HPC applications. The `-affinity` flag prevents the operating system from moving the MPI processes between cores. They run exclusively on the core that they started on.

Additional references