TCP拥塞控制-拥塞避免算法：cwnd≥ssthresh还是仅cwnd>ssthresh时启动？

Great question—this is a super common point of confusion when unpacking TCP's congestion control logic. Let's cut straight to the answer first: 拥塞避免算法启动的条件是当cwnd（拥塞窗口）大于等于ssthresh（慢启动阈值）时，而不是仅当cwnd严格大于ssthresh。

Let me walk through the standard TCP Tahoe/Reno behavior to make this clear:

• When a connection starts, TCP is in slow start mode: cwnd starts at 1, and grows exponentially each RTT (round-trip time) (basically, it doubles each round trip as ACKs come in).
• The second cwnd hits the value of ssthresh, TCP immediately switches to congestion avoidance. Instead of exponential growth, cwnd now increases linearly—only adding 1 per full RTT (usually after receiving ACKs for an entire window of data).
• If congestion is detected later (like via lost packets), ssthresh gets updated to half of the current cwnd, and cwnd is reset (Tahoe sets it back to 1; Reno cuts it in half and enters fast recovery). TCP will then go back to slow start until cwnd climbs back up to the new ssthresh, at which point it switches right back to congestion avoidance.

Here's a quick concrete example: Suppose your initial ssthresh is set to 10.

• When cwnd is 9: Still in slow start, so it'll keep growing exponentially (next RTT might take it to 18, depending on ACK timing).
• When cwnd hits 10: Boom—congestion avoidance starts. Now each RTT, cwnd will tick up by 1: 10 → 11 → 12, and so on, until congestion is detected.

A quick note: While most standard implementations follow this cwnd ≥ ssthresh rule, a few custom TCP variants might tweak this slightly, but this is the universally accepted standard behavior for core TCP stacks.

内容的提问来源于stack exchange，提问作者Walter