##Discussed##

  • BOSCO: plugin that takes jobs from OSG queue, puts into Slurm cluster, juggles the two, etc.
  • API for OpenStack migration Check migration and live-migration apis from OpenStack to fulfill the purpose of a low prioirty VM migration.
  • Build a story:
    • backfill job on node A, high priority job on B, new job comes along, throws out A, then B finishes.
    • so migrate backfill job to A and start it up on a differnet physical node.
  • Network piece: connecting Kaizen to MIT.
    • We just have to work through it, do all the necessary routing.
    • Over tunnels is fine, just have to make it work.
    • We’re not as concerned about security, this is a proof of concept demonstration.
  • Can do:
    • create two tunnels, network across (which scales to a tunnel from every node to every node…).
    • Not very practical. Create a single tunnel with internal routing.
    • Using multiple tunnel method, we can “say” we added a virtual node to a real cluster.
      • Method: slurm controller needs a tunnel which connects it to virtual node.
      • Also need a tunnel to connect virtual node to salt master.
  • Should make a list of steps to try to connect virtual node to main cluster.
    • What resources do we need to test this on engage1?
      • Physical controller on engage1
      • Salt master (which could be on Kaizen)
      • virtual node on Kaizen
      • Would need two way tunnel between controller and virtual node.
  • Steps:
    • Provision a node
    • Create tunnels (both on MIT end and NEU end, controller -> node and vice versa)
    • MIT shares munge key
    • MIT edit slurm.conf
    • Other files can be shared by “FUSE”: https://github.com/libfuse/libfuse
    • This gets us the majority of the way there!