8.3.1. How Job Data Files are Handled in EKM and RSM

This section describes:

8.3.1.1. The Data Management Process

A data management process is used to both stage and retrieve files for jobs submitted to RSM. It is a Java process that runs on the cluster submit host as the same user who submits the RSM job.

When a user starts a job in EKM, the user’s data management process writes job data to the job data directory, which is specified in the jobSourceRootPath setting in the ekm.xml file. The cluster submit host accesses this directory when managing the job.

In order for the data management process to execute properly, the job data directory must meet the following requirements:

  • The directory must exist before EKM is deployed and users begin running jobs, and must be accessible by the cluster submit host.

  • All users who will be submitting jobs must have read/write permission on this directory.

For more information about the jobSourceRootPath setting, see Specifying Remote Process Policies (<remoteProcess>).

To see how the data management process works, see the example in The Creation and Structure of Job Directories.

8.3.1.2. The Creation and Structure of Job Directories

When users set up a batch job in EKM, they have the option of choosing a working directory for the job. If they do not choose one, EKM will stage job files in the job data directory. This directory is specified in the jobSourceRootPath setting in the ekm.xml file.

If a shared job data directory is being used, EKM will create a workspace directory under the job data directory (if not already present), followed by a username directory, and time stamp directory.

As an example in Windows:

  1. In the ekm.xml file, the jobSourceRootPath is set to \\HEADNODE\HPCTemp. This is a UNC path to a shared directory named HPCTemp on the cluster submit host. This translates to C:\HPCTemp on that machine. This is also the cluster staging directory.

  2. John Smith (with user name jsmith) is signed in to the Default workspace. When John runs a job, jsmith’s data management process creates the following directory for the job:

    C:\HPCTemp\Default\jsmith\20_May_2015_10.39.18_AM
  3. When John selects input files for the job, the files are uploaded to the 20_May_2015_10.39.18_AM directory. The cluster submit host accesses this directory to manage the job. Since the working directory resides under the cluster staging directory (C:\HPCTemp), the files do not need to be transferred to the cluster staging directory.

  4. Every time John starts a new job, a new subdirectory named with the job’s time stamp will be created under the C:\HPCTemp\Default\jsmith directory.

  5. When another user, Susan Thomas, starts a job, a directory will be created for her job under the job data directory in a similar way:

    C:\HPCTemp\Default\sthomas\20_May_2015_11.40.26_AM

In a Linux configuration, if home directories are being used instead of a shared job directory, only a time stamp directory will be created under the job data directory, as shown in the following example:

/net/server/home/jsmith/jobRoot/20_May_2015_10.39.18_AM

For information about specifying the jobSourceRootPath setting, see Specifying Remote Process Policies (<remoteProcess>).


Note:  If using a shared job data directory, it is recommended that you create the workspace subdirectory up front, and ensure that all users who will be submitting jobs have read/write permission on that subdirectory. Additional steps may be needed depending on your security requirements. For more information, see Access and Permission Requirements of the Job Data Directory.


8.3.1.3. Access and Permission Requirements of the Job Data Directory

Before job submission begins, you must create the job data directory that you have specified in the jobSourceRootPath setting, and ensure that it is accessible to all users who will be submitting jobs. EKM will create workspace, user, and time stamp subdirectories as described in The Creation and Structure of Job Directories. There are important considerations to be made regarding the job data directory and its subdirectories. If you are working in Linux, additional steps may be required depending on the directory that you have chosen, and the level of security that you desire.

In Windows:

  • The job data directory must be a shared directory, specified by a UNC path in the jobSourceRootPath setting.

  • Make sure that this directory is open to all users who will be running RSM jobs.

  • It is recommended that you create the workspace subdirectory up front, and ensure that all users have read/write permission on this directory, so that each user’s data management process will be able to create a user directory for that user under the workspace directory. For example, if your job data directory path is \\HEADNODE\HPCTemp, and users will be submitting jobs from the Default workspace, create a Default directory under the HPCTemp directory (\\HEADNODE\HPCTemp\Default), and set appropriate permissions on the Default directory.

In Linux:

  • The use of home directories (via a relative path) is the recommended way of handling users’ job data in Linux. Home directories provide each user with guaranteed access to his or her own job data, while preventing unwanted access by other users. The home directory is accessible by RSM, and the permission requirements are satisfied because users have read/write access on their own home directories.

  • You may choose to use a shared job directory (via an absolute path) instead of home directories. This may be less secure, however, because users could potentially have access to each other’s data. If you do choose to use a shared job directory, it is recommended that you create the workspace subdirectory up front, and ensure that all users have read/write access on this subdirectory. Although not necessary, for tighter security you may also want to create user subdirectories up front and grant each user full read/write permissions on his or her user subdirectory rather than the workspace directory. Keep in mind, however, that if you remove read/write permissions on the workspace directory, you will need to create user subdirectories for any new users who will be added to EKM later on, and set up appropriate permissions on those directories.

    It is important to ensure that all users have at least access (execute) permission on all parent folders in the chain, right up to the top level. Consider the path /someshare/folder1/jobRoot/Default. If you granted users read/write permissions on the jobRoot and Default folders, but did not grant them access permission on folder1, job submission would fail for those users.

For information about specifying the jobSourceRootPath setting, see Specifying Remote Process Policies (<remoteProcess>).


Release 18.2 - © ANSYS, Inc. All rights reserved.