Approach to Increasing Parallelism

• Starting point

  • Assume a Giant-free network stack
  • Select an interesting workload
  • What are remaining source of contention?
  • Where is CPU-intensive activity serialized in a single thread – leading to unbalanced CPU use?

• Identify natural boundaries in processing

  • Protocol hand-offs, layer hand-offs, etc
  • Carefully consider ordering considerations

• Weigh trade-offs, look for amortization

  • Context switches are expensive
  • Locks are expensive