none
Internal mechanism of adding preconfigured compute node RRS feed

  • Question

  • During the process of adding preconfigured compute node by assigning a compute node template, HPC pack first discovering the configuration and interfaces on the relevant node computer, retrieving the disk volume information, then it starts creating an administrators group, then it will adding and removing users from the group.

    My Question is:

    What’s the purpose of this creation of administrators group, and why to add and remove users from this group?  Would anybody please kindly explain the internals behind this process?  I believe many people would be interested in this. Thanks a lot.


    Feng Wang
    Friday, May 20, 2011 1:44 PM

Answers

  • Hi Feng Wang,

    The original behavior of our product was to simply synchronize the head node's Administrators and Users groups with that of each of your compute nodes.  By doing this, the product ensures that any cluster admin has enough priveleges to ensure proper cluster-wide admin operations and all users can submit/run jobs on any compute node.

    Starting with V3 (HPC Pack 2008 R2) we've refined this behavior by replicating new user groups called "HPCAdminMirror" and "HPCUsers" as a subset of the default system "Administrators" and "Users" groups.  This allows you to have more system Admins/Users on a particular compute node that are not necessarily HPC cluster admins/users.  I.e. you must be a member of these managed "HPC.." groups to perform cluster-wide actions (e.g. HPCAdminMirror can 'clusrun', while HPCUsers can only submit/monitor their own jobs).

    Regards,
    --Brian


    • Edited by Brian Broker Monday, June 20, 2011 10:05 AM font issue + clarification
    • Marked as answer by 51gaoxing Friday, August 5, 2011 7:28 PM
    Monday, June 20, 2011 10:01 AM

All replies

  • Hi Feng Wang,

    The original behavior of our product was to simply synchronize the head node's Administrators and Users groups with that of each of your compute nodes.  By doing this, the product ensures that any cluster admin has enough priveleges to ensure proper cluster-wide admin operations and all users can submit/run jobs on any compute node.

    Starting with V3 (HPC Pack 2008 R2) we've refined this behavior by replicating new user groups called "HPCAdminMirror" and "HPCUsers" as a subset of the default system "Administrators" and "Users" groups.  This allows you to have more system Admins/Users on a particular compute node that are not necessarily HPC cluster admins/users.  I.e. you must be a member of these managed "HPC.." groups to perform cluster-wide actions (e.g. HPCAdminMirror can 'clusrun', while HPCUsers can only submit/monitor their own jobs).

    Regards,
    --Brian


    • Edited by Brian Broker Monday, June 20, 2011 10:05 AM font issue + clarification
    • Marked as answer by 51gaoxing Friday, August 5, 2011 7:28 PM
    Monday, June 20, 2011 10:01 AM
  • Hi Brian,

    Thank you for taking your time illuminating me along with this thread.  Hopefully the HPC group @ Microsoft can publish something HPC internals or something. A lot of developers will like that, don't you think so?

    Have a great day.

     


    Feng Wang
    Friday, August 5, 2011 7:32 PM