TCP AIMD

The AIMD approach to congestion control involves increasing/decreasing the size of packet windows to control congestion.

• Senders can increase sending rate until packet loss (i.e. congestion occurs)
• Decrease sending rate on loss events
• Additive increase
  • Increase sending rate by 1 maximum segment size until loss is detected (cwnd)
  • Max window size ⇒ number of segments sent at once
• Multiplicative Decrease
  • Sending rate is cut in half on loss detected by triplicate duplicate (ACK) - Reno
  • Sending rate is cut to 1 MSS (max segment size) when loss is detected by timeout
    • Worst case ⇒ all packets in a window were dropped

AIMD has sawtooth behavior:

AIMD is a distributed, asynchronous algorithm that has been shown to:

• Optimize congested flow rates network wide
• Have desirable stabaility properties

Details:

• TCP sender behavior (roughly) is to send cwnd bytes, wait RTT for ACKs, then send more bytes
• TCP rate $\approx \frac{\text{cwnd}}{RTT}$ bytes/second
• TCP sender will limit transmission
  • LastByteSent - LastByteAcked ≤ cwnd
• cwnd is dynamically adjusted in response to observed network congestion

TCP slow start is used for the initial increase in packet send rate:

• When connection begins, increase rate exponentially until the first loss event occurs
  • cwnd starts at 1 MSS
  • cwnd is doubled every RTT
    • Implemented by incrementing cwnd for every ACK received