Principles of the TCP Protocol

The Transmission Control Protocol (TCP) is the core protocol that implements end-to-end reliable data transmission in internet communication. Unlike UDP's "best effort" strategy, TCP ensures packets arrive at their destination accurately, in order, and completely through a series of complex algorithms and state machine designs. Understanding how TCP works is crucial for building highly available network applications. This article delves into the three core mechanisms of TCP: connection establishment, flow control, and congestion control.

Three-Way Handshake and Four-Way Wave: Dynamic Connection Management

TCP uses a connection-oriented communication method, which means a logical pathway must be established before data transmission begins, and resources must be released after transmission ends. This process ensures synchronization between both parties through the "three-way handshake" and "four-way wave."

1.3 Analysis of the Three-Way Handshake Principle

The initiator (client) and receiver (server) need to exchange three data packets to confirm consistency in their network status and sequence numbers:

  • First SYN Packet: The client sends a segment containing the SYN flag, carrying an initial sequence number IS_N. This requests connection establishment.
  • Second ACK + SYN Packet: Upon receiving it, the server replies, sets its own synchronization state, acknowledges the client's request (ACK), and simultaneously initiates its own connection request (SYN). This step confirms mutual intent.
  • Third ACK Packet: The client sends another ACK segment to acknowledge the server's sequence number. At this point, the three-way handshake is complete, and the connection formally enters the ESTABLISHED state to prepare for data transmission.

Skipping any of these steps can lead to the receiver being unable to correctly parse subsequent packets or falling into an infinite retransmission loop. For example, without the second SYN-ACK, the server cannot know it needs to wait for the client's ACK; without the third ACK, the connection will remain in a half-open state forever.

2. Necessity of the Four-Way Wave

After data transmission is complete, TCP uses the "four-way wave" to gracefully close the connection:

  • First FIN: The active initiator sends the FIN flag, indicating that its own data has been transmitted and requesting disconnection.
  • Second ACK: Upon receiving it, the passive receiver replies with an ACK. At this point, it may still have pending data to transmit, so it does not immediately sever the connection.
  • Third FIN: Once all data transmission is complete on the server side, it actively sends its own FIN flag to notify the client that it is ready to close.
  • Fourth ACK: The client finally acknowledges the server's disconnection request.

It is worth noting that TCP's half-duplex nature dictates this asymmetric operation flow is necessary because both parties need to independently judge whether they are ready to release buffer resources.

Synergistic Operation of Flow Control and Congestion Control

To ensure reasonable utilization of network resources and packet integrity, TCP includes two key regulation mechanisms internally: the sliding window protocol addresses receiver processing capability limitations (flow control), while algorithms like slow start and fast retransmit optimize for overall network load (congestion control).

1. Window-Based Flow Control

To prevent the sender's speed from being too fast and causing receiver buffer overflow, TCP introduces the "advertised window" (rwnd) field. Every packet header carries the value of this field, informing the other party of the current available buffer space size.

  • Dynamic Adjustment: When the receiver's data fills up, it sets rwnd to 0; as processing completes and memory is freed, rwnd gradually increases.
  • Sliding Window Mechanism: The sender decides how many unacknowledged packets it can transmit continuously based on the received maximum window value (i.e., the minimum of the congestion window cwnd and the advertised window). This dynamic matching effectively avoids the situation where the receiver discards data due to a full buffer.

2. Four Phases of Congestion Control

When packets are transmitted over the network, they may encounter packet loss caused by router queue overflow. TCP balances the sending rate using four algorithms:

  • Slow Start: In the early stages of connection establishment, cwnd grows exponentially to quickly probe the upper limit of network bandwidth. Once a timeout or three duplicate ACKs are detected, it enters the next phase.
  • Congestion Avoidance: When approaching a bottleneck, cwnd increases linearly, cautiously probing the limits of network load.
  • Fast Retransmit and Fast Recovery: If the sender continuously receives three duplicate ACKs, it indicates that packet loss occurred in the middle without a timeout. At this point, it immediately enters "fast recovery" state, resetting cwnd to one less than the number of received duplicate ACKs (minus the amount of already acknowledged data), and either restarting slow start or continuing the congestion avoidance cycle, significantly reducing retransmission latency.

3. Practical Application Scenario Example

Assume a video streaming application transmits high-definition frames via TCP: if network fluctuations cause packet loss, TCP's fast recovery mechanism can quickly adjust the sending rate; meanwhile, the speed of the receiver's decoder determines the size of rwnd, thereby preventing stuttering caused by frontend rendering falling behind. This dual-control strategy makes TCP the preferred protocol for long-lived connections and large data volume transmission.

In summary, through rigorous state machine design and adaptive regulation algorithms, TCP builds a solid foundation for reliable communication in complex network environments. Mastering its handshake process and control mechanisms is a critical step for developers optimizing application performance.