Make Widom's cluster operational, including adding new hardware. 2 phases.

Phase 1

Get Widom's current nodes operational for researchers.

Add new nodes, when they arrive (~end of April 2014).

Strategy

Borrow hardware from ChemIT to serve as a dedicated head node until ChemIT able to convert one of Widom's compute nodes into a headnode.

Timing?

Phase 2

When?

Take off-line one of Widom's nodes for about a month and convert it to a head node.

  • Decision required: Use a new node, or an older (~2 year old) node?
    • Oliver's current recommendation: Use a new node. Pros/ cons?
  • This leaves 3 compute nodes in production, on ChemIT's loaner head node.

About a month later:

Take another compute node (old or new?) off-line to test with the newly created head node. Needed for duration of test.

  • This leaves 2 compute nodes in production, on ChemIT's loaner head node.

Take a 3rd compute node off-line (opposite age of node already being tested) to test with the newly created head node (~1/2 day?)

  • This leaves 1 compute node in production, on ChemIT's loaner head node.

Once testing is completed, wipe out data from newly created head node. Turn off production system (using ChemIT's loaner head node). Transfer data to new head node. Turn on new head node as new production system.

  • ChemIT's loaner head node hardware returned to ChemIT.
  • No labels