Congestion Control Algorithm

Computer Network & Network Security System — Transport Layer, NEC licence examination syllabus (Nepal Engineering Council).

Congestion Control: keeping the network from collapsing

Flow control asks whether the receiver can cope. Congestion control asks whether the network can — and nobody sends you an answer.

What congestion actually is

Congestion happens where more traffic arrives at a router than it can forward. The excess queues, the queue fills, and further arrivals are dropped. The damaging part is what happens next: every sender whose packet was dropped retransmits, which adds more traffic to an already overloaded router.

That feedback loop is why congestion collapses rather than degrading. Beyond a certain load, increasing offered traffic decreases the amount successfully delivered, because a growing share of what the network carries is retransmissions of things it already dropped. Every congestion control mechanism exists to keep the network below that point.

TCP's answer: a window that probes and retreats

No router tells a sender to slow down. TCP infers congestion from packet loss, on the reasoning that in a wired network loss almost always means a full queue rather than a corrupted bit.

TCP congestion window (cwnd), in segments: SLOW START — exponential growth start cwnd = 1 every acknowledgement received → cwnd += 1 effect: cwnd DOUBLES each round trip 1 → 2 → 4 → 8 → 16 → 32 … "Slow" is historical: it starts small, but it is the FASTEST-growing phase. It ends at the slow-start threshold (ssthresh). CONGESTION AVOIDANCE — linear growth above ssthresh: cwnd += 1 per ROUND TRIP, not per ack 32 → 33 → 34 → 35 … Probing gently for more capacity now that the earlier doubling has got close to the limit. ON LOSS ssthresh = cwnd / 2 remember roughly where it broke cwnd = 1 (timeout) and start again or = ssthresh (fast recovery, on duplicate acks)

💡 The name "slow start" is the most-missed point in this topic. It is not the slow phase — it is the exponential one. It is called slow because it begins at one segment rather than blasting a full window immediately, which is what earlier implementations did and what caused the congestion collapses that motivated the design.

Shaping traffic: leaky bucket and token bucket

Both smooth a bursty sender into something the network can carry, and the difference between them is exactly whether bursts are allowed.

LEAKY BUCKET Packets pour into a bucket with a hole in the bottom. Output rate is CONSTANT, whatever the input does. Bucket full → new packets are discarded. → perfectly smooth output, no bursts ever TOKEN BUCKET Tokens accumulate in a bucket at a fixed rate. A packet may be sent only by spending a token. Tokens saved during quiet periods can be spent at once. → average rate is capped, but SHORT BURSTS ARE ALLOWED

The practical distinction: a leaky bucket makes an idle sender waste its allowance, because the hole drains whether or not anything is queued. A token bucket lets that sender save it up and transmit a burst later, which suits real traffic — applications are naturally bursty, and forbidding bursts entirely wastes capacity that nobody else was using.

Commonly confused

Congestion control and flow control. Congestion is the network; flow is the receiver. The receiver advertises a window; the network does not.

Slow start being slow. It is the exponential phase.

Leaky bucket and token bucket. Leaky enforces a constant rate. Token enforces an average rate while permitting bursts.

Loss meaning congestion. True on wired links, and a genuine weakness on wireless, where loss is often interference — TCP then slows down for no reason.

Rapid revision

Congestion collapses because dropped packets are retransmitted, adding load to an overloaded network.

Slow start doubles cwnd per round trip until ssthresh; congestion avoidance then adds one per round trip.

On loss: halve ssthresh, and restart at 1 after a timeout.

Leaky bucket = constant output. Token bucket = average rate with bursts permitted.

Syllabus points

Congestion: causes and effects
TCP congestion control: slow start, congestion avoidance (AIMD)
Leaky bucket & token bucket

Create a free account to tick topics off, take notes as you read, watch the video lessons and get a day-by-day study plan built around your exam date.