3.7 - TCP Congestion Control | FHU - Computer Networks

Summary

This script discusses TCP's congestion control mechanisms, focusing on how TCP perceives congestion through packet loss and timeouts. It explains TCP's strategies for limiting the send rate using the congestion window (cwnd) and how it adapts its rate with additive increase and multiplicative decrease algorithms. The script outlines TCP's three modes: slow start, congestion avoidance, and fast recovery, illustrating how TCP increases its sending rate until a loss event triggers a reduction, aiming to transmit as fast as possible without causing congestion.

Takeaways

• 📶 **Congestion Perception**: TCP perceives congestion through implicit feedback, monitoring for lost packets as an indication of network congestion.
• 🚦 **Congestion Control**: TCP limits sending rate by adjusting the congestion window (cwnd), which is the number of unacknowledged bytes allowed in the pipeline.
• 🔄 **Adaptive Algorithms**: TCP uses algorithms like additive increase and multiplicative decrease (AIMD) to adapt its sending rate in response to perceived congestion.
• 📈 **Slow Start and Congestion Avoidance**: TCP starts with an exponential increase in the congestion window during 'slow start' and shifts to a linear increase in 'congestion avoidance' after reaching a certain threshold.
• 🔁 **Loss Events Handling**: Upon a loss event, TCP enters 'fast recovery' mode, cutting the congestion window size in half, and then increases it more conservatively.
• 🔄 **Timeout vs. Duplicate ACKs**: A timeout results in a drastic reduction of the congestion window to one MSS, while receiving three duplicate ACKs halves the congestion window.
• 🌐 **Network Utilization**: TCP aims to transmit data as fast as possible without causing congestion, balancing between utilizing available bandwidth and avoiding network overload.
• 🔄 **Congestion Window Dynamics**: The congestion window size is adjusted based on the round-trip time and the acknowledgments received, affecting the sender's rate.
• 📉 **Rate Adjustment**: TCP increases its rate additively upon receiving ACKs and decreases it multiplicatively upon detecting loss events.
• 🔧 **TCP States**: TCP operates in different states (slow start, congestion avoidance, fast recovery) to manage its congestion control effectively.

Q & A

How does TCP perceive congestion?

TCP perceives congestion through an end-to-end basis by listening for lost packets or failure events at the sender and receiver.

What is the quick and easy answer to how TCP knows if congestion is going on?

The quick answer is that TCP infers congestion by monitoring for packet losses or delays in acknowledgments.

How does TCP limit the send rate?

TCP limits the send rate by controlling the congestion window (cwnd), which is the number of unacknowledged bytes that are allowed to be in the pipeline.

What is the purpose of the congestion window (cwnd) in TCP?

The congestion window (cwnd) is used to limit the sender to the minimum of the receive window or the congestion window, effectively controlling the rate of data sent into the network.

How does TCP change its rate in response to congestion?

TCP changes its rate using algorithms like additive increase, multiplicative decrease (AIMD), which adjust the congestion window size in response to success and loss events.

What are the two main modes of TCP congestion control?

The two main modes of TCP congestion control are slow start and congestion avoidance.

How does the congestion window increase in slow start mode?

In slow start mode, the congestion window increases exponentially, doubling every round-trip time until a loss event occurs or until it reaches the slow start threshold.

What happens when the congestion window reaches the slow start threshold?

When the congestion window reaches the slow start threshold, TCP switches from slow start mode to congestion avoidance mode, where the window increases linearly.

How does TCP respond to a timeout, which is a loss event?

In response to a timeout, TCP drastically reduces the congestion window to one maximum segment size and returns to slow start mode.

What is the difference between TCP Tahoe and TCP Reno in handling loss events?

TCP Tahoe resets the congestion window to 1 upon a loss event, while TCP Reno cuts the congestion window in half and continues in congestion avoidance mode.

What is the role of fast recovery in TCP congestion control?

Fast recovery is an intermediate mode between slow start and congestion avoidance. It increases the congestion window more conservatively than slow start but is not as drastic as a timeout, cutting the window in half upon entering it.