What Is Network Congestion
In simple terms, network congestion means that packets sent over the network are lost in transit or arrive with higher latency because the link is overloaded. Whether you see loss or delay depends on how routers, switches, and other devices along the path are configured. Generally, when the network is congested, packets are first buffered at routers or switches; if that exceeds their capacity, packets are dropped.
When Does Network Congestion Happen
You can think of a network link like shipping a package. You drop off a parcel at a local courier station A. If few people are sending packages, you barely wait. If many people are there, you queue up. When sending network packets, if many users are already on the access network and bandwidth is limited, congestion happens.
After the package leaves station A, it goes to a regional hub B. B receives parcels from many cities. Packages from A may be few, but other regions may send a lot, so B can also get overwhelmed. When B cannot keep up, there are two options: stack parcels in the warehouse (buffering and delay), or throw some away (packet loss).
If B is not congested, the package reaches destination hub C. C receives parcels from all over the country and may also hit peak seasons like Singles' Day or 618, overflowing capacity. Your package is then likely delayed again. From C, parcels are dispatched to local stations in each county, and those stations may also have long pickup queues, so congestion can happen there too.
So, analogous to the courier network, congestion can occur at every stage of the Internet. As long as everyone keeps sending packets, even very large bandwidth can be filled.
How Do You Know Congestion Has Occurred
This is exactly what various congestion control algorithms try to figure out. As the sender, once a packet leaves, you do not know what happens on the path—which routers it passes through, which links it uses, and so on.
So how do you judge congestion?
One approach is delay-based estimation. The sender attaches a timestamp; the receiver collects statistics on received packets and echoes the send timestamp back. The receiver can then measure end-to-end delay. If delay exceeds a threshold, congestion is likely. Using the courier analogy again: if a package normally arrives in 3 days but takes 4 or 5, the network is mildly congested; if it takes a week or half a month, you probably hit Singles' Day and congestion is severe.
Another approach is loss-based detection. If some packets are lost, the receiver only gets part of the stream and reports back: you sent 10 packets, I only got 5, and I missed 1, 4, 7, 8, and 9. That also strongly suggests congestion.
In other words, the courier station threw away 5 of your 10 parcels because it could not handle the load.
What Should You Do When Congestion Happens
Normally, when congestion occurs, senders should reduce the amount of traffic. If everyone cuts back appropriately—for example to 80% of the previous rate—the network load drops, and both loss and delay improve.
Fairness and Ethics
What if someone does not voluntarily lower their send rate when bandwidth is congested? For example, if congestion causes 10% packet loss, sending more packets could increase the number that arrive and preserve communication quality.
In theory that works for one sender, but if everyone does it, total traffic rises again. Eventually everyone suffers worse loss and a worse network experience.
Like courier stations: when they are overloaded and discard 10% of parcels, you send 10% more and the number delivered stays the same. If everyone thinks the same way and sends 10% more, the hub overflows and decides: there are too many parcels, workers cannot keep up, let's discard 20%.
Now everyone has 20% loss and nobody is happy.
So under congestion, everyone should yield appropriately and send less. One person sending a bit more is usually fine; when many do it, nobody gets a good connection.
Congestion Control in Operating Systems
TCP congestion control is built into the system. Most users do not change it casually. Its strategy is to back off when congestion occurs so everyone can still get online with rough fairness.
If congestion algorithm behavior were fully under each user's control—send as much as you want with no coordination—the Internet would break down. Like a slightly jammed highway: you speed up to cut in line; everyone fears being stuck and speeds up to cut in too; eventually the whole road blocks and nobody moves.
City driving restrictions work on the same principle.
The Carrier's Stance
Carriers are relatively friendly to TCP because TCP has flow control. UDP often has no congestion control, or implements it separately, which makes it easier for someone to send more packets to grab bandwidth and increase delivery.
That kind of bandwidth hogging may be tolerable during off-peak hours. But at peak times, carriers may limit UDP to keep TCP stable—for example by dropping packets.
Then some people find UDP quality poor. The carrier may simply be throttling you: normal users can barely get reliable service, and you are still trying to squeeze in more packets.
Knowing how carriers and firewalls behave, should you send more UDP packets to fight their intentional drops? Better not! Carriers will increase drop rates, often in bursts, and your QoS will suffer badly.
The Firewall's Stance
Companies usually have network operations teams who are wary of UDP bandwidth grabbing, so some enterprises are unfriendly to UDP. UDP is also associated with many network attacks, so it is even less welcome.
Yet UDP is a foundational protocol for real-time audio and video. That leads to cases where the corporate network is fine but calls are choppy or impossible. Hence RTP over TCP.
Pure Packet-Loss Networks
Is there packet loss that is not caused by carriers, firewalls, or other "abnormal" factors?
Yes, but it is rare.
Why? Think of the Internet as a plumbing network. Large-scale leaks are unlikely; when they happen, they are fixed quickly.
So if a user often sees 10%–20% loss, that is usually unlikely, because ordinary web browsing would already be broken and they would likely call the carrier for repair.
Afterword
In the end, the essence of network congestion is everyone competing for limited bandwidth, which leads to overload. The symptoms are delay plus packet loss.
This article was written stream-of-consciousness and may be a bit wordy.
