Analysis of TCP Fast Open by Google

Google will publish their work in improving the WEB application response time in the ACM CONEXT meeting in December this year, using three handshakes “TCP Fast Open” when using three handshakes by modifying the TCP protocol. Although Paper is released two days ago, the relevant RFC draft submitted IETF as early as March 2011, and a UPDATE was carried out two weeks, here is DIFF.

For the content of the TCP Fast Open program, a friend of Taobao has been interpreted according to the draft of the RFC. I will not let it be, interested in friends can see Paper or RFC. I just want to discuss the application prospect of this thing.

Since there is no deep understanding of the background, I believe that many people have tried to do a similar job, but I think similar work should have not received large-scale applications. For TCP protocols that have already been formed for a long time, it is difficult to modify its desire, because changed the underlying thing means a lot of trouble.

But if you are willing to pay a price, there is a premise that there is no sufficient benefit. The benefits given by TFO are: When RTT (Round Trip Time) is relatively low, the client page load time is optimized approximately 4% to 5%; the longer RTT, the greater the benefits, the average is about 25%.

Google TCP Fast Open Evaluation

In addition to page loading, TFO has also brought some benefits to the server side. Since each request saves an RTT, the CPU consumption of the server is also reduced accordingly. The data given in Paper is a number of transaction per second rose from 2876.4 to 3548.7.

Although most of the time in Paper emphasizes TFO’s significant acceleration effect on web page, I think even if TFO can become an Internet standard, its current state is still a long distance, so it is impossible in the short term. To the mainstream Internet world. But this does not mean that it doesn’t have the opportunity, and the illegitishers of the young brother, there may be two directions in its promotion application:

1. Move the Internet. The RTT of the mobile Internet is now much larger than the traditional Internet (frequently managed, data support), so that an RTT saving effect cannot be ignored; additional mobile Internet terminal operating system is diversified, unlike a desktop terminal system. Single. Google has a very important Android, Baidu also plans to launch its own “easy platform”. These Internet companies have motivated to improve the user experience of mobile users to access their own websites.

2. Internet Enterprise Data Center. Although the data center is delayed, it is still attractive to reduce the benefits of RTT, and attractically reduce the calculation capacity waste and increasing throughput. Coupled with many companies used to be customized open source operating systems and customized network libraries, the price of the upgrade is not so high. If I am the person in charge of the enterprise infrastructure, I think I will carefully consider this program. …