So what's next ?

• Relaxing functional equivalence necessary

  • Do we always have to determine the "best" task to run ? When is "close" good enough ?
    • PLS: pick first task in highest PLS queue.
      • Limit number of priority levels
    • e.g. MQ: scan subset of remote cpus.
  • NUMA
    • Differentiated affinity intra/inter node
  • Throughput more important than strict priority at high end ?

Notes: