> [TOC]

# 1、TCP 历史及其设计哲学

## 1.1、TCP/IP 的前身 ARPA：NCP 协议

![image-20220819103419076](assets/image-20220819103419076.png)

## 1.2、TCP/IP 协议发展

![image-20220819103534060](assets/image-20220819103534060.png)

## 1.3、TCPv4 协议分层后的互联网世界

![image-20220819103643226](assets/image-20220819103643226.png)

## 1.4、TCP/IP 的七个设计理念

![image-20220819103731990](assets/image-20220819103731990.png)

# 2、TCP 解决了哪些问题？

## 2.1、TCP 的作用

![image-20220819103925515](assets/image-20220819103925515.png)

## 2.2、TCP协议的分层

![image-20220819104250668](assets/image-20220819104250668.png)

## 2.3、报文头部

![image-20220819104444880](assets/image-20220819104444880.png)

![image-20220819104539556](assets/image-20220819104539556.png)

## 2.4、TCP 协议特点

![image-20220819104557349](assets/image-20220819104557349.png)

# 3、TCP 报文格式

## 3.1、消息传输的核心要素

![image-20220819104907273](assets/image-20220819104907273.png)

## 3.2、IP头部

![image-20220819105028413](assets/image-20220819105028413.png)

![image-20220819105414479](assets/image-20220819105414479.png)

## 3.3、UDP 头部

![image-20220819105049681](assets/image-20220819105049681.png)

![image-20220819105434419](assets/image-20220819105434419.png)

## 3.4、TCP 协议的任务

![image-20220819105137234](assets/image-20220819105137234.png)

## 3.5、如何标识一个连接？

![image-20220819105206106](assets/image-20220819105206106.png)

## 3.6、TCP Segment 报文段

![image-20220819105254952](assets/image-20220819105254952.png)

![image-20220819105355131](assets/image-20220819105355131.png)

* options选项

![image-20220819105532495](assets/image-20220819105532495.png)

# 4、三次握手建立连接

## 4.1、握手的目标

![image-20220819105816744](assets/image-20220819105816744.png)

## 4.2、三次握手

![image-20220819110855076](assets/image-20220819110855076.png)

## 4.3、报文格式

![image-20220819110950677](assets/image-20220819110950677.png)

![image-20220819111000165](assets/image-20220819111000165.png)

![image-20220819111006994](assets/image-20220819111006994.png)

# 5、三次握手状态变迁

## 5.1、三次握手流程

![image-20220819112615675](assets/image-20220819112615675.png)

## 5.3、netstat 命令查看TCP 状态

![image-20220819112843131](assets/image-20220819112843131.png)

## 5.2、两端同时发送SYN：双方使用固定源端口且同时建连接

![image-20220819112853238](assets/image-20220819112853238.png)

# 6、三次握手中的性能优化与安全问题

## 6.1、服务器三次握手流程示例

![image-20220819121223337](assets/image-20220819121223337.png)

## 6.2、超时时间与缓冲队列（调整SYN队列大小）

![image-20220819121317717](assets/image-20220819121317717.png)

## 6.3、Fast Open 降低时延

![image-20220819121417253](assets/image-20220819121417253.png)

![image-20220819121444425](assets/image-20220819121444425.png)

## 6.4、如何应对 SYN 攻击？

![image-20220819121503161](assets/image-20220819121503161.png)

![image-20220819121549058](assets/image-20220819121549058.png)

## 6.5、TCP_DEFER_ACCEPT

* 直到收到data分组再激活nginx功能

![image-20220819121648544](assets/image-20220819121648544.png)

# 7、数据传输与MSS 分段

## 7.1、TCP 应用层编程示例

![image-20220819121947777](assets/image-20220819121947777.png)

## 7.2、TCP 流的操作

![image-20220819122002481](assets/image-20220819122002481.png)

![image-20220819122008552](assets/image-20220819122008552.png)

* IP层分层损耗过大

## 7.3、MSS：Max Segment Size（不包含头部大小）

![image-20220819122137556](assets/image-20220819122137556.png)

## 7.4、TCP 握手常用选项

![image-20220819122235037](assets/image-20220819122235037.png)

# 8、重传与确认

## 8.1、报文有可能丢失

![image-20220819122533560](assets/image-20220819122533560.png)

## 8.2、解决方法

![image-20220819122540118](assets/image-20220819122540118.png)

* 串行执行
* 效率低下

![image-20220819122605119](assets/image-20220819122605119.png)

* 限制发送

## 8.3、存在问题（针对每一个字节）

![image-20220819122719074](assets/image-20220819122719074.png)

![image-20220819122725584](assets/image-20220819122725584.png)

## 8.4、序列号复用

![image-20220819123023018](assets/image-20220819123023018.png)

![image-20220819123042085](assets/image-20220819123042085.png)

* 解决方法：采用时间戳

## 8.5、BDP 网络中的问题

![image-20220819123223221](assets/image-20220819123223221.png)

# 9、RTO 重传定时器的计算

## 9.1、如何测量 RTT？

![image-20220819135413904](assets/image-20220819135413904-16608884542752.png)

![image-20220819135432532](assets/image-20220819135432532.png)

## 9.2、TCB

**Socket包含两部分，一个是IP地址，一个是端口号。**同一个设备可以对应一个IP地址，但不同的管道用不同的端口号区分，于是同一个设备发送给其他不同设备的信息就不会产生混乱。在同一时刻，设备可能会产生多种数据需要分发给不同的设备，为了确保数据能够正确分发，TCP协议用一种叫做**TCB（Transmission Control Block，传输控制块）**的数据结构把发给不同设备的数据封装起来。

## 9.3、RTO（ Retransmission TimeOut ）应当设多大？

![image-20220819140107392](assets/image-20220819140107392.png)

## 9.4、RTO 应当更平滑

![image-20220819140155305](assets/image-20220819140155305.png)

## 9.5、追踪 RTT 方差（linux实际方案）

![image-20220819140302454](assets/image-20220819140302454.png)

# 10、滑动窗口：发送窗口与接收窗口

## 10.1、滑动窗口：发送窗口快照

![image-20220819140426261](assets/image-20220819140426261.png)

## 10.2、可用窗口

![image-20220819140538028](assets/image-20220819140538028.png)

![image-20220819140622055](assets/image-20220819140622055.png)

![image-20220819140633201](assets/image-20220819140633201.png)

## 10.3、发送窗口

![image-20220819140708300](assets/image-20220819140708300.png)

## 10.4、接受窗口

![image-20220819140804930](assets/image-20220819140804930.png)

# 11、窗口的滑动与流量控制

## 11.1、示例

![image-20220819141127964](assets/image-20220819141127964.png)

## 11.2、客户端消息的发送

![image-20220819142823430](assets/image-20220819142823430.png)

## 11.3、服务器消息的发送

![image-20220819142838668](assets/image-20220819142838668.png)

# 12、操作系统缓冲区与滑动窗口的关系

## 12.1、窗口与缓存

![image-20220819143103155](assets/image-20220819143103155.png)

## 12.2、收缩窗口导致的丢包

![image-20220819143813603](assets/image-20220819143813603.png)

## 12.3、合适的窗口大小

![image-20220819144026237](assets/image-20220819144026237.png)

## 12.4、Linux下调整接收窗口与应用缓存

![image-20220819144051325](assets/image-20220819144051325.png)

![image-20220819144058827](assets/image-20220819144058827.png)

# 13、如何减少小报文提高网络效率？

## 13.1、SWS(Silly Window syndrome)糊涂窗口综合症

![image-20220819144154909](assets/image-20220819144154909.png)

## 13.2、SWS 避免算法

![image-20220819144329292](assets/image-20220819144329292.png)

* nagle算法等待ACK时候累计小报文
* 如果报文超过MSS，则不论是否收到ACK，必须立即发送

## 13.3、TCP delayed acknowledgment 延迟确认

![image-20220819144456771](assets/image-20220819144456771.png)

## 13.4、Nagle VS delayed ACK

![image-20220819144632429](assets/image-20220819144632429.png)

# 14、拥塞控制（1）：慢启动

## 14.1、全局思考：拥塞控制

![image-20220819145021860](assets/image-20220819145021860.png)

## 14.2、定义

![image-20220819145141426](assets/image-20220819145141426.png)

* 通告窗口：实际上就是接受窗口

## 14.3、慢启动的初始窗口

![image-20220819145257235](assets/image-20220819145257235.png)

# 15、拥塞控制（2）：拥塞避免

## 15.1、定义

![image-20220819145620822](assets/image-20220819145620822.png)

## 15.2、慢启动与拥塞控制

![image-20220819145739606](assets/image-20220819145739606.png)

# 16、拥塞控制（3）：快速重传与快速恢复

## 16.1、为何会接收到一个失序数据段？

![image-20220819150016060](assets/image-20220819150016060.png)

## 16.2、快速重传（RFC2581）

![image-20220819150139111](assets/image-20220819150139111.png)

## 16.3、注意事项、

![image-20220819150425355](assets/image-20220819150425355.png)

![image-20220819150410274](assets/image-20220819150410274.png)

## 16.4、快速恢复（RFC2581）

![image-20220819150522131](assets/image-20220819150522131.png)

# 17、SACK 与选择性重传算法

## 17.1、仅重传丢失段

![image-20220819151255001](assets/image-20220819151255001.png)

## 17.2、重传所有段

![image-20220819151447238](assets/image-20220819151447238.png)

## 17.3、SACK：TCP Selective Acknowledgment

![image-20220819151508048](assets/image-20220819151508048.png)

![image-20220819151521134](assets/image-20220819151521134.png)

# 18、四次握手关闭连接

## 18.1、关闭连接

![image-20220819151812534](assets/image-20220819151812534.png)

## 18.2、两端同时关闭连接

![image-20220819152208472](assets/image-20220819152208472.png)

## 18.3、TCP 状态机

![image-20220819152233264](assets/image-20220819152233264.png)

# 19、优化关闭连接时的TIME-WAIT状态

## 19.1、TIME-WAIT状态过短或者不存在会怎么样？

![image-20220819152436497](assets/image-20220819152436497.png)

## 19.2、linux下TIME_WAIT优化：tcp_tw_reuse

![image-20220819152636177](assets/image-20220819152636177.png)

## 19.3、TIME_WAIT 优化

![image-20220819152700842](assets/image-20220819152700842.png)

## 19.4、RST 复位报文（直接关闭连接）

![image-20220819152744767](assets/image-20220819152744767.png)

# 20、keepalive 、校验和及带外数据

## 20.1、TCP 的 Keep-Alive 功能

![image-20220819152850211](assets/image-20220819152850211.png)

## 20.2、违反分层原则的校验和

![image-20220819152902603](assets/image-20220819152902603.png)

## 20.3、应用调整 TCP 发送数据的时机

![image-20220819152927259](assets/image-20220819152927259.png)

![image-20220819152935358](assets/image-20220819152935358.png)

# 21、面向字节流的TCP 连接如何多路复用？

## 21.1、Multiplexing 多路复用

![image-20220819153123437](assets/image-20220819153123437.png)

## 21.2、HTTP2：TCP 连接之上的多路复用

![image-20220819153227180](assets/image-20220819153227180.png)

## 21.3、非阻塞 socket：同时处理多个TCP连接

![image-20220819153239936](assets/image-20220819153239936.png)

## 21.4、epoll

![image-20220819153258192](assets/image-20220819153258192.png)

![image-20220819153350185](assets/image-20220819153350185.png)

![image-20220819153403993](assets/image-20220819153403993.png)

# 22、四层负载均衡可以做什么？

![image-20220819153549110](assets/image-20220819153549110.png)

![image-20220819153556558](assets/image-20220819153556558.png)

![image-20220819153605829](assets/image-20220819153605829.png)

![image-20220819153619358](assets/image-20220819153619358.png)

![image-20220819153626455](assets/image-20220819153626455.png)