Mitigating Locking Overhead

• Amortize cost of locking

  • Avoid multiple lock operations where possible
  • Amortize cost of locking over multiple packets

• Coalesce/reduce number of locks

  • Excessive granularity will increase overhead
  • Combining across layers can avoid lock operations necessitated by to lock order

• Serialization “by thread”

  • Execution of threads is serialized

• Serialization “by CPU”

  • Use of per-CPU data structures and pinning/critical sections