Finally addressing a long-term problem. . .

Problem:

• User demand is not uniform — e.g., conference/publication deadlines.

Solution:

• Burst HPC cluster into AWS Spot Market? Not a good match.
• Condor:
  • Aggregate compute resource wherever we can find it.
  • Natural match for "spare" cycles in AWS, Azure, etc. —
    • cost-effective.
  • Almost invisible to users!