TCP Congestion Control: Slow Start
What you'll learn: How TCP cautiously probes available network bandwidth by starting small and growing exponentially.
Why Slow Start?
When a TCP connection begins, the sender doesn't know how much data the network can handle. Blasting packets at full speed would overwhelm routers and cause congestion collapse. Instead, TCP uses slow start to carefully test the waters.
How Slow Start Works
TCP maintains a congestion window (cwnd) that limits how many unacknowledged segments can be in flight. Slow start begins conservatively:
- Initial window: Start with a small cwnd (typically 1-10 segments)
- Exponential growth: For each ACK received, increase cwnd by one segment
- Result: The window doubles every round-trip time (RTT)
If cwnd = 1, you send 1 segment. When its ACK arrives, cwnd = 2. Send 2 segments. When both ACKs arrive, cwnd = 4. Then 8, 16, 32...
This exponential ramp-up quickly discovers available capacity without immediately flooding the network.
When Does Slow Start End?
Slow start continues until:
- Packet loss detected (timeout or duplicate ACKs) — signals congestion
- Slow start threshold (ssthresh) reached — switches to a more cautious linear growth phase
Think of it like merging onto a highway: you start slowly, accelerate quickly until you sense traffic conditions, then adjust your speed more carefully.
Real-World Impact
Slow start is why web pages with many small connections can feel slow — each connection restarts from cwnd = 1. Modern optimizations like TCP Fast Open help, but slow start remains fundamental to preventing network collapse.
Key Takeaway: TCP's slow start exponentially increases the congestion window from a conservative initial value, probing network capacity safely before ramping up to full speed.