Imagine a highway with thousands of cars. If too many cars enter the highway at the same time, traffic jams start to happen.
A computer network can have the same problem. If too much data is sent into the network at once, routers and links can become overloaded. Packets may be delayed or dropped.
This situation is called network congestion.
TCP Congestion Control is the set of mechanisms TCP uses to control its sending rate so that it does not overwhelm the network.
In simple words:
TCP Congestion Control prevents a sender from putting too much traffic into the network at once.
What Is Network Congestion?
Network congestion happens when the amount of traffic entering a network is greater than what the network can handle.
For example:
Many Senders
↓
Router
↓
[Overloaded]
↓
Packets delayed/dropped
When congestion occurs:
Packets may be delayed
Packets may be dropped
Queues at routers can grow
Network performance can decrease
TCP tries to detect signs of congestion and adjust its sending rate.
Why Does TCP Need Congestion Control?
Suppose a server can send data at a very high speed, but the network path between the server and client can handle only a limited amount of traffic.
If the server keeps sending at maximum speed:
Sender
↓
Too much traffic
↓
Network becomes congested
↓
Packets are dropped
↓
Retransmissions
↓
More traffic
This can make the situation even worse.
TCP therefore increases its sending rate carefully and reduces it when it detects signs of congestion.
Congestion Window (cwnd)
One of the most important concepts in TCP congestion control is the congestion window, commonly written as cwnd.
The congestion window limits how much data TCP can have in flight based on its estimate of network capacity.
There are actually two important limits:
Receive Window (rwnd) → Protects the receiver
Congestion Window (cwnd) → Helps protect the network
TCP's effective sending window is constrained by both.
A simplified relationship is:
Effective Sending Window
≈
minimum(rwnd, cwnd)
So even if the receiver can accept a lot of data, TCP may still limit transmission because of network congestion.
Main TCP Congestion Control Mechanisms
Traditional TCP congestion control is commonly explained using four major ideas:
Slow Start
Congestion Avoidance
Fast Retransmit
Fast Recovery
Different TCP implementations use different algorithms and improvements, but these concepts are important for understanding the basics.
1. Slow Start
Despite its name, Slow Start can increase the congestion window quite rapidly.
When a TCP connection begins, the sender does not know how much traffic the network can handle.
So it starts with a relatively small congestion window and increases it as acknowledgments arrive.
Conceptually:
1 → 2 → 4 → 8 → 16 → ...
This behavior is commonly described as exponential growth, although the exact behavior depends on the TCP implementation and conditions.
The purpose is to quickly discover available network capacity without immediately sending a huge amount of traffic.
2. Congestion Avoidance
Once the congestion window reaches a certain threshold, TCP generally changes to a more cautious growth strategy.
This phase is called Congestion Avoidance.
Instead of rapidly increasing the window, TCP increases it more gradually.
A simplified example:
16 → 17 → 18 → 19 → 20 → ...
This helps TCP use available bandwidth while reducing the risk of causing congestion.
3. Fast Retransmit
TCP can detect packet loss without always waiting for a retransmission timer to expire.
If a receiver receives later data while an earlier segment is missing, it can send duplicate acknowledgments for the missing portion.
For example:
Sender → Segment 1
Sender → Segment 2
Sender → Segment 3
Sender → Segment 4
Receiver receives:
1, 3, 4
Receiver → Duplicate ACK
Receiver → Duplicate ACK
Multiple duplicate acknowledgments can indicate that a segment was probably lost.
Fast Retransmit allows TCP to retransmit the missing segment sooner instead of waiting for the retransmission timeout.
4. Fast Recovery
After detecting packet loss through duplicate acknowledgments, some TCP congestion-control algorithms reduce their sending rate and then recover without returning all the way to the initial slow-start behavior.
This is known as Fast Recovery.
The exact behavior depends on the TCP congestion-control algorithm being used.
The general idea is:
Normal Transmission
↓
Packet Loss Detected
↓
Reduce Sending Rate
↓
Recover Gradually
What Happens When Congestion Is Detected?
TCP treats packet loss and other signals as evidence that the network may be congested.
A simplified view is:
Increase Sending Rate
↓
Network Congestion?
/ \
No Yes
↓ ↓
Increase Reduce
further sending rate
↓
Increase gradually
This continuous adjustment allows TCP to adapt to changing network conditions.
AIMD
A common concept associated with traditional TCP congestion control is AIMD, which stands for:
Additive Increase, Multiplicative Decrease
Additive Increase
When the network appears healthy, TCP gradually increases its congestion window.
Multiplicative Decrease
When congestion is detected, TCP reduces its congestion window significantly.
A simplified example:
10 → 11 → 12 → 13 → 14
↓
Congestion
↓
7
↓
8 → 9 → 10
The exact values and behavior depend on the TCP algorithm.
AIMD helps TCP find a balance between using available bandwidth and avoiding excessive congestion.
Flow Control vs Congestion Control
This is one of the most important distinctions in TCP.
Flow Control
Flow control protects the receiver.
It asks:
"Can the receiving device handle this much data?"
TCP uses the receive window (rwnd) for this purpose.
Congestion Control
Congestion control protects the network.
It asks:
"Can the network path handle this much traffic?"
TCP uses the congestion window (cwnd) and congestion-control algorithms.
Flow Control | Congestion Control |
|---|---|
Protects receiver | Protects network |
Based on receiver capacity | Based on network conditions |
Uses receive window | Uses congestion window |
Prevents receiver overload | Helps prevent network overload |
Simple Example
Imagine you are downloading a large file.
At first, TCP sends data cautiously:
Small amount
↓
Network
↓
ACKs received
↓
Send more
As long as the network appears healthy, TCP increases its sending rate.
If packet loss suggests congestion:
Packet Loss
↓
Congestion Detected
↓
Reduce Sending Rate
↓
Recover Gradually
So TCP is constantly trying to find a good sending rate for the current network conditions.
Modern TCP Congestion Control
It is important to know that TCP congestion control is not one single algorithm.
Different operating systems and network environments can use different algorithms, such as:
Reno
New Reno
CUBIC
BBR
They use different strategies to estimate network capacity and respond to congestion.
For beginners, however, the most important concepts are congestion window, slow start, congestion avoidance, packet loss, and rate reduction.
Conclusion
TCP Congestion Control is used to prevent TCP from overwhelming the network with too much traffic.
TCP adjusts its sending rate based on signals about network conditions.
The key concepts are:
cwnd → Controls the amount of data TCP can have in flight based on congestion conditions
Slow Start → Rapidly increases the sending window when starting or recovering
Congestion Avoidance → Increases the sending rate more cautiously
Fast Retransmit → Quickly retransmits likely-lost data
Fast Recovery → Recovers after certain loss events without restarting from the beginning
AIMD → Traditional principle of increasing gradually and decreasing significantly after congestion
The easiest way to remember it is:
Flow Control → Don't overwhelm the receiver.
Congestion Control → Don't overwhelm the network.