15 - Congestion Control

congestion

a network condition where demand for resources exceeds available capacity, leading to delays and packet loss.

cwnd

the congestion window, a sender-controlled limit on the amount of data that can be in transit.

rwnd

the receive window, a receiver-controlled limit on how much data the sender can transmit.

slow start

a TCP phase where the congestion window increases exponentially to probe available bandwidth.

ssthresh

the slow start threshold, defining the point at which TCP switches from slow start to congestion avoidance.

congestion avoidance

a TCP phase where the congestion window grows linearly to cautiously probe for bandwidth.

aimd

Additive Increase, Multiplicative Decrease; a mechanism to increase bandwidth usage gradually and reduce it aggressively upon detecting congestion.

duplicate ack

repeated acknowledgment for the same data, often indicating packet loss or reordering.

fast recovery

a TCP mechanism triggered by three duplicate ACKs, allowing quick recovery from isolated losses.

timeout

a condition where the sender retransmits data after a period without acknowledgment, indicating severe congestion or multiple losses.

ecn

Explicit Congestion Notification, a network-assisted mechanism where routers signal congestion to endpoints.

bottleneck

a network point where traffic demand exceeds capacity, causing delays and packet drops.

fairness

the principle ensuring equitable bandwidth distribution among competing network flows.

retransmission

resending data that was lost or not acknowledged, aimed at ensuring reliable delivery.

sawtooth pattern

the characteristic behavior of TCP congestion control, with periodic increases and decreases in the congestion window.

available bandwidth

the portion of a link’s capacity that can be used without causing congestion.

packet drop

the loss of a data packet due to full buffers or other network issues.

round trip time (rtt)

the time it takes for a packet to travel to the receiver and for the acknowledgment to return.

multiplicative decrease

reducing the congestion window size significantly (typically halving) after detecting congestion.

additive increase

incrementing the congestion window size slowly to probe for additional bandwidth.

tcp tahoe

a TCP variant implementing slow start, congestion avoidance, and fast retransmit but not fast recovery.

tcp reno

a TCP variant enhancing Tahoe by adding fast recovery for better performance during isolated losses.

tcp vegas

a TCP variant that detects congestion early by monitoring RTTs, aiming for better efficiency.

tcp cubic

a TCP variant optimized for high-speed and long-delay networks, using a cubic function for window growth.

tcp westwood

a TCP variant that uses bandwidth estimation to adjust congestion control parameters dynamically.