Background: SMPng Motivation

• 2.x – No SMP

  • Unable to use additional processors

• 3.x/4.X SMP – “Giant Kernel Lock”

  • Kernel protected by one Giant spin lock
    • No kernel parallelism with SMP
    • Processes burn CPU while waiting
  • Limited preemption resulted in latency issues

• Urgent need: scale well with >1 processor

  • Hyper-threading, dual-core becoming a reality