TCP acceleration principle and BBR enhanced version for Debian/Ubuntu

TCP Acceleration Principles and BBR Enhanced Edition for Debian/Ubuntu

TCP acceleration

1Basic content 2TCP transparent proxy 3TCP acceleration technology Bilateral TCP acceleration Unilateral TCP acceleration Commercial TCP acceleration 1 Basic content Edit TCP acceleration refers to using a set of optimization technologies to make TCP more performant in transmitting data over the Internet (compared to standard TCP); at the same time, there is no need to modify specific applications.

Common TCP acceleration technologies include TCP transparent proxy and TCP offloading engine.

2 TCP Transparent Proxy Edit The working principle of TCP transparent proxy is as follows:

The transparent proxy interacts with both ends of the TCP connection respectively, thus dividing the end-to-end TCP control into several parts, so that different optimizations can be carried out according to the packet loss and delay of these parts, thereby improving TCP performance.

The core idea of ​​the TCP accelerator is to use a transparent proxy to terminate the connection at one end of the TCP and then reinitiate a connection to the other end of the TCP. In this way, data packets at both ends are cached on the TCP accelerators at both ends, and data transmission between TCP accelerators is controlled by the TCP accelerator.

3 TCP acceleration technology editor Currently, the mainstream TCP acceleration technologies mainly include bilateral TCP optimization and unilateral TCP optimization.

Bilateral TCP acceleration

Bilateral TCP optimization, as the name suggests, is to deploy hardware devices or install software at both ends of the TCP connection. TCP transparent proxies work at both ends of the TCP connection, proxying the connections at both ends. The two proxies are usually connected through UDP or other Custom protocols work.

In actual use, both ends of the TCP protocol and software or hardware devices are in a LAN, and the WAN link between the two transparent proxy devices usually has certain packet loss and delay, which will cause TCP performance to degrade. Therefore, between these two transparent proxies, the protocol is usually converted to UDP protocol or other custom protocols. These protocols themselves can be completely controlled according to their own requirements to achieve the effect of improving TCP performance; at the same time, bilateral TCP acceleration can also be introduced Compression, caching and other technologies further improve TCP performance.

Bilateral TCP optimization is more suitable for companies with multiple branches. In this case, both ends of the TCP connection are usually easier to control, and hardware devices or software clients can be installed more easily.

Unilateral TCP acceleration

Unilateral TCP acceleration means that software or equipment can be deployed on only one end of TCP to achieve the goal of improving TCP performance.

A basic requirement for unilateral TCP acceleration is that the protocol passed through the transparent proxy must be the TCP protocol (including 5-tuples and various states of TCP). A transparent proxy for unilateral TCP acceleration should run a TCP on the WAN side that is compatible with standard TCP and has improved performance. The vast majority of unilateral TCP acceleration accelerates TCP by improving TCP's congestion control algorithm, such as TCP Vegas, CUBIC, FastTCP, Zeta-TCP, etc.

Compared with bilateral TCP, unilateral TCP optimization is more adaptable and flexible. For example, as long as TCP acceleration is performed on the server side, all clients accessing this server will benefit, and the client does not need to install any software or deploy hardware equipment. In this way, it is more suitable for situations where the access target of the server is not fixed, for example, a certain server is accessed by a large number of Internet users.

However, unilateral TCP acceleration cannot directly implement functions such as compression and caching. If these functions are to be realized, bilateral deployment is also required.

================================================ ===

Compared with Ruisu, BBR does not occupy resources and has better network optimization. The modified version below is more suitable for the network environment in mainland China than the original version. It works perfectly on Debian8, Debian9, and Ubuntu16.04. Under ssh:

wget --no-check-certificate -qO 'BBR.sh' 'https://zhujiwiki.com/usr/uploads/2017/08/BBR.sh' && chmod a+x BBR.sh && bash BBR.sh - f

Script description: The module will be recompiled during execution. By default, the module is loaded automatically at boot. Module name:tcp_bbr_powered Modprobe tcp_bbr_powered command can be used to load modules. Executable lsmod |grep ‘bbr_powered’ If the result is not empty, the module is loaded successfully. Execute sysctl -w net.ipv4.tcp_congestion_control=bbr_powered to use this module. The above is just an explanation, just use the one-click script.

Recommended site search: Zhengzhou server, foreign virtual host, independent ip virtual host, domain name, com domain name registration, domain name registration query, shopex virtual host, Korean independent server, Alibaba Cloud free virtual host, registration number,