Mapped drives in HPC Pack 2016 RRS feed

  • Question

  • What's currently the best way to handle mapped drives?  To date (in HPC Pack 2012) our users have used a node prep task full of "net use" statements, but I'm not convinced it's not causing some issues.  I saw some very old discussion about always using UNC, but was wondering if there's something better with the latest release since we're slowly moving to it.
    Tuesday, August 15, 2017 7:02 PM

All replies

  • In windows compute nodes, mapping a network drive is a per user logon session persisted. When the job is finished/completed, the mapped drive will be gone away. And as it is per-session, other user's job won't be able to access. And as on domain joined compute nodes, UNC is also widely used for jobs.

    In linux compute nodes, we have "execution filter" to allow mounting user UNC path, the document around this will come out following months. If you need this, let me know.

    Qiufang Shi

    Wednesday, August 16, 2017 1:36 AM
  • Thanks.  Each user runs their own work so it's not an issue that it's per-user, though they generally all have access to the same drives anyway.  Most of their codes are based on drive letters instead of UNC; they can easily change it but many of the DFS UNC paths are pretty long.

    Here's an example of the node prep task used:

    net use ^* /delete /y & net use Z: "\\\shares\dfs1" & net use Y: "\\\shares\dfs2" & net use X: "\\\shares\dfs3"

    If it's considered best practice to use UNC I can get them to change, though if there should be nothing wrong with what's done now we can keep using that.  I'm also curious if anything changed with this from 2012 to 2016; our first test job in 2016 didn't write out where it was supposed to.

    Workstation nodes are all Windows 10, for what it's worth.  No Linux compute nodes for now though that's possible in the future.


    Wednesday, August 16, 2017 1:59 PM