TCP Congestion Control: Congestion Avoidance
What you'll learn: How TCP switches from exponential to linear growth of its congestion window to gently probe for available bandwidth without overwhelming the network.
The Problem After Slow Start
You've learned that TCP Slow Start doubles the congestion window (cwnd) exponentially until it hits the slow start threshold (ssthresh). But what happens next? If TCP kept doubling, it would quickly cause congestion again. That's where Congestion Avoidance comes in.
Linear Increase: Playing It Safe
Once cwnd reaches ssthresh, TCP switches gears. Instead of doubling every round-trip time (RTT), it increases the window size linearly—typically by adding just one maximum segment size (MSS) per RTT.
Think of it like this: Slow Start is like flooring the gas pedal to accelerate onto the highway. Congestion Avoidance is like easing off and gently accelerating to find the speed limit without getting a ticket (packet loss).
The Additive Increase Rule
For each RTT where all segments are acknowledged successfully:
cwnd = cwnd + MSS
This cautious, additive increase allows TCP to probe for additional bandwidth without triggering congestion. The sender carefully tests whether the network can handle slightly more traffic.
Why Linear?
Linear growth prevents aggressive overshooting. If the network becomes congested (detected by packet loss or timeout), TCP reduces cwnd dramatically and resets ssthresh, then the cycle repeats: Slow Start up to the new threshold, then Congestion Avoidance resumes.
This creates the classic "sawtooth" pattern of TCP throughput over time—gradual increase, sharp drop on congestion, repeat.
Key Takeaway: TCP Congestion Avoidance uses linear (additive) increase of the congestion window after reaching the slow start threshold, gently probing for available bandwidth to avoid overwhelming the network.