Congestion Control

• Backpressure
  • A node quite a way from the source begins to experience congestion.
  • Packets from the preceding node begin to experience delay, causing congestion on that node.
  • This continues back to the source.
  • This can happen
    • Just because it does
    • To selected links
    • To selected sources
    • To selected applications
  • But it requires more intelligence as we go up the list
  • This requires a network that can do hop-by hop flow control
  • Which most don't do.
• Choke Packet
  • This is an ICMP source quench.
  • It should be delivered to the source to tell it to cut back.
  • Stallings calls this a "crude technique"
• Implicit Congestion Signaling
  • This relies on sources to detect that packet delays are increasing.
  • Or to detect increased packet loss
  • Take this as a signal of trouble and reduce transmission rates.
  • This will fix the problem.
  • This works in connectionless/datagram networks.
  • TCP performs at this level.
  • This requires no support from the lower levels of the network.
  • This is also called end-to-end congestion control
• Explicit Congestion Signaling
  • On connection oriented networks.
  • All nodes upstream, back to the source, can be notified to slow down.
    • This is done with header information in return values (like acks)
    • Or through explicit messages.
  • Or the down stream nodes can be notified
    • In this case, flow control should be passed on to a higher level (TCP)
    • Or the destination can send information back in acks.
    • This can be done by changing packet headers as well.