Running Parallel Programs

This section describes how to run a parallel program. In PTP, this is called "running a job".

Previous releases of PTP required setting up a Resource Manager for each job scheduler/runtime environment before a job could be launched. In PTP 6.0 and later this is not handled separately, and instead a Target System Configuration is specified on the Run Configuration page. Further interaction to set up the parallel job is highly dependent on which TSC is selected.

PTP is able to launch parallel applications that use a variety of programming models (e.g. MPI, OpenMP, UPC, etc.) provided that the Target System Configuration supports the programming model. It is also possible to launch multiple copies of a sequential program so that they execute simultaneously.

Eclipse uses a run configuration (also known as a launch configuration) to encapsulate all the settings and arguments necessary to launch a job normally or under the control of a debugger. A run configuration is created using the Run > Run Configurations... menu.

This section only describes how to launch a parallel program. A description of how to monitor the application launch is provided in Monitoring Jobs and Systems. A description of how to debug an application is provided in Parallel Debugging.

Topics include:

  1. Creating a Parallel Application run configuration
  2. Completing the Resources tab - selecting the target system configuration
  3. Completing the Application tab
  4. Completing the Arguments tab
  5. Completing the Environment tab
  6. Completing the Synchronize tab
  7. Running the application

Step 1: Create a Parallel Application run configuration

  1. To set up a run configuration, open the Run Configurations dialog using Run > Run Configurations...


    or click the run icon's context menu on the tool bar (small triangle next to the icon) and select Run Configurations....


  2. Create a new Parallel Application run configuration by clicking Parallel Application and then clicking on the New button icon.



  3. Enter a name for the configuration in the Name: field. Here we use the same name as the project. If the project was selected in the Project Explorer, it may be already filled in here.


Step 2: Complete the Resources tab

The Resources tab is used to specify the resources required for the job execution. After selecting the target system configuration, the contents of this tab will vary for different target configurations, and your installation may have added others.

PTP supported Target System Configurations

PTP includes a number of target configurations for different systems that support various parallel runtime systems and job schedulers. The following table lists the common configurations along with links to additional help documentation.

Note that the associated runtime or job scheduler needs to be available on the target system prior to using PTP:

Target Configuration Description
Grid Engine-Generic-Batch Batch execution using Grid Engine
IBM LoadLeveler Generic IBM LoadLeveler batch systems
IBM LoadLeveler (Blue Gene) IBM LoadLeveler on Blue Gene systems
IBM Parallel Environment Interactive execution using IBM Parallel Environment
IBM Platform LSF Batch execution on IBM Platform LSF systems
IBM Platform MPI IBM Platform MPI
MPICH2-Generic-Interactive Interactive execution on MPICH2 systems
Open MPI-Generic-Interactive Interactive execution on Open MPI systems
PBS-Generic-Interactive Interactive job execution using PBS batch systems
PBS-BGP-Batch Blue Gene/P systems using
PBS-BGQ-Batch Blue Gene/Q systems using PBS
SLURM-Generic-Batch Generic SLURM batch systems
SLURM-ALPS-Generic-Batch Combination of SLURM and ALPS as batch system
SLURM-BGP-Batch Blue Gene/P systems using SLURM
SLURM-BGQ-Batch Blue Gene/Q systems using SLURM
PBS-Generic-Batch Generic PBS batch systems
Remote-Generic-Interactive Simple (non-parallel) interactive execution on remote systems
Torque-Generic-Batch Generic TORQUE batch systems
Torque-Generic-Interactive Interactive job execution using Torque batch systems
de.fz-juelich.judge.torque.batch Running batch jobs on the JSC system JUDGE
de.fz-juelich.juropa.torque.batch Running batch jobs on the JSC system JUROPA
de.fz-juelich.juqueen.ll_bg Running batch jobs on the JSC system JUQUEEN

  1. Select a Target System Configuration you wish to use from the drop-down.

     

     

    Note that if you type a lone character (or a few) into the selection box, the list of target system configurations will be limited to those that start with the character.

  2. Select the connection to use for the target system. If you have previously configured a synchronized or remote project, then a connection should be available, and may be already be selected.
    (To launch on the local machine, select Local for the connection type.)
    You can create a new connection here by clicking the New... button.

     

  3. Once you select the connection, you may be prompted to open the connection if it is not already open. The contents of the Resources tab will then populate with controls that allow you to configure the necessary launch information for the target system. This area usually (depending on the target system configuration) includes further tabs to segment the other information you can provide. Fill in the information required by the target system configuration.

Step 3: Complete the Application tab

The Application tab is used to specify the application (executable) to be launched.

  1. Ensure that the correct Parallel Project is selected. This is the project containing the source and binaries for the application.


  2. Select the Application Program (executable) you wish to launch. If the target system connection points to a remote target, then this means the path of the executable on the remote target. Otherwise, for a connection pointing to the local machine, it should be the path to the executable within the project directory.

    The Browse... button can be used to select the executable file.


  3. If you are running the application on a remote machine and the project was built locally, then you will have to copy the executable to the remote machine first. For this purpose, you may enable Copy executable from local filesystem and fill Path to the local file with the path to the executable within the project directory. This configures the launcher to copy the executable automatically to the remote target before starting the run/debug session. If not enabled, then you will need to copy the executable manually to the remote target.



  4. If you wish to have the output from all processes in a single output console, ensure that the Display output from all processes in a console view option is checked.


Step 4: Complete the Arguments tab

The Arguments tab is used to specify any command-line arguments require by the parallel application and to modify the working directory.

  1. Enter any arguments required by the application in the Program arguments field. This is a free form text field that will be passed to the application verbatim.


  2. Select a different working directory if desired. The working directory is set just prior to the application launch. Note that for a remote launch, this will be a directory on the remote machine.


Step 5: Complete the Environment tab

The Environment tab is used to specify any environment variables require by the parallel application or in order to run the application. The tab also provides the option to append the environment variables to the native environment, or to replace the native environment with the variables specified in the tab.

  1. Use the New... button to create a new environment variable.


  2. Use the Select... button to import environment variables.


  3. Use the Edit... button to edit an environment variables.


  4. Use the Remove... button to remove an environment variables.



Step 6: Complete the Synchronize tab

The Synchronize tab is used in scenarios where the connection refers to a remote target. The tab specifies rules that describe files or directories that need to be copied (uploaded) to the remote target before starting the parallel application execution. Typically, this feature is used to upload dependencies (like dynamic libraries) or input data for test cases.

Note that this may not be necessary for Synchronized Projects, in which files are copied back and forth to the remote system independently of the launching mechanism.

This tab also allows to specify rules describing files on the remote host to be retrieved back (downloaded) to the local host after the parallel application finishes execution, as output data produced by the parallel application execution.

  1. If you wish to enable file transfer to the remote target, enable the Upload rules enabled or Download rules enabled according which direction of file transfer is required.



  2. In order to specify files to be copied from the local system to the remote target (uploaded), click on New upload rule. An upload rule lists a collection of files or directories that shall be copied into a single directory.


    If Use directory from launch configuration is enabled, then the listed files or directories are copied to the working directory specified on the launch configuration. At the moment, this is equivalent to same directory that contains the executable for the parallel application. If not enabled, they you must specify on Remote directory a path where the files or directories will be placed. The path may be absolute or relative (to the working directory).


    In the list underneath, Selected file(s), add the files or directories that shall be copied. Click on File(s) to select one or more individual files from anywhere in the local file system. The Directory button works similarly to select a directory from the local file system. Directories are copied recursively. The Workspace button allows easily choosing files from the current workspace.


    On the bottom of the dialog, you may wish to set attributes to be applied to the copied files on the remote target. The options Readonly or Executable sets all copied files to read-only or executable on the remote host. If Preserve time attributes is enabled, then the remote copy will assume the same time stamp then the local file, otherwise, the time stamp will be the time the file was copied. Also, if the file already exists on the remote target, you may choose how to react in the If file already exists dropdown.




  3. In order to specify files to be copied from the remote target to the local system (downloaded) after the parallel application finishes, click on New download rule.


    An download rule lists a collection of files or directories that shall be copied into a single local directory. Its options are very similar to the upload rule.



Step 7: Launching the application

The launch configuration is now complete. Select the Run button (in the bottom right corner of the Run Configurations dialog to launch the job. Switch to the System Monitoring perspective to view the status of the job and inspect the program output.

Back to Top | Back to Table of Contents