Background and current situation of domestic CDN acceleration

sigcomm An interesting paper ：
https://conferences.sigcomm.org/events/apnet2021/papers/apnet2021-4.pdf
The macro state from the micro perspective . Roughly speaking ：

• although BBR Overwhelming advantage leads people to use BBR Replace CUBIC, But as the BBR Increase in deployment ratio , The income will be smaller and smaller , Finally, each cc The proportion of the algorithm is balanced , Continue deployment BBR Your motivation will disappear , The future Internet is a cc Heterogeneous networks .

This is an analysis of the typical phenomenon of diminishing marginal benefits .

Same perspective , Various domestic Internet manufacturers CDN Why can't the transmission acceleration race stop ？ CDN Itself is the acceleration technology of the Internet , Right again CDN Is it meaningful to accelerate ？

Yesterday I posted a circle of friends ：

Recommend an article about BBR And game theory :( That is, the opening quotation ). Combined with the transmission acceleration of domestic Internet manufacturers , Finally stand outside the circle and watch , No domestic manufacturer has such a perspective , None of them , No manufacturer will consider the investment income ratio and fairness , All manufacturers are radical transmission on a case by case basis , Simply focusing on technology but not so proficient in technology , In short, there are no observable sustainable quantitative benefits , This is a small loss is a big gain , Lose the game less than who loses , And the wages of workers in this industry are not low , Personally, it doesn't look like a loss , The loss is the company . But there is a certain threshold in this industry , And it is not easy to see benefits ( It may have been squeezed out of soup ..), So I can't roll it up at the personal level , But the company is involved ！ Don't you see that every traditional Internet company has a position of transmission protocol optimization in recruitment ？ What do you expect ？ Expect an overwhelming advantage ？ Shouldn't you focus on the business itself ？ perhaps , Think about how to optimize IDC The Internet , Good cloud primordial , This is the business ！ Internet companies IDC No transmission , Various incast None of them solved , Solve these problems more , Benefit the country, the people, the company, the manager and himself .

Reflect on domestic CDN Manufacturers' transmission acceleration , Also tends to be balanced , But the arms race continues , Not for better , I'm afraid it will be worse if I don't do it .

CDN It's a good technology , No, CDN, Our access to the Internet will be slow , Again diaodiaodiao Your site is also unable to load balance , The more diaodiaodiao The more serious the site path convergence , The more congestion , yes CDN Decoupling load balancing .CDN Acceleration technology in itself , Yes CDN How significant is it to accelerate itself ？

As early as 2012～2014 year , Net residence ,APPEX etc. CDN Peripheral technology manufacturers have led the transmission acceleration , tencent , Ali , Bytes to beat , Kwai, these large and traditional Internet manufacturers, cannot rely on external technology for a long time to consume huge costs , But increasing the transmission rate is just needed , Sooner or later, the transmission acceleration , Join the competition .

In the early era of mobile Internet , Acceleration is very simple , As long as you send a few more packets per round, it will be effective , It's easy to understand , When limited resources are not overloaded , Basically, those who come here have a share , Even whoever can grab it is his , This is the same as breaking through the Kanto and grabbing land , Shenzhen Development , In fact, the same principle applies to the early purchase and settlement of houses in Pudong, Shanghai , But this kind of cheap will be more and more difficult to take . In fact, the manufacturers who tasted the benefits of transmission acceleration in the early stage are nethome ,APPEX etc. .

I once had a point , If the total resources are not expanded , The competition of diminishing marginal benefits will sooner or later make everyone give up and invest again , Reach equilibrium . Therefore, transmission optimization will become increasingly unpromising , The effect of diminishing marginal benefits is that workers' income decreases .

But I didn't think , The competition between manufacturers cannot converge to a stable state . let me put it another way , As long as one company starts , Other homes have to start up , Finally, everyone hangs in the air to balance , Need to continuously invest resources , Who let go first , Who falls first .

BBR and CUBIC Reaching Nash equilibrium will stop deploying BBR, The heterogeneous state will remain stable , The transmission acceleration competition of Internet manufacturers is continuous , The input is the worker's salary .

technically , Radical transmission has long been unable to bring observable additional benefits , But this behavior has to continue .3 People hold 3 Mouth pot , No one dares to let go first , If you let go, you will be smashed .

Someone has to raise the pot below , Therefore, workers' income will not be affected by diminishing marginal benefits , The important thing is that the pot can't fall down . This is a typical involution competition in which less loss is more gain . What is wasted is the money and food of the company , But if the company does not invest , There will be more losses .

People ask whether the guarantee of Internet fairness depends on the goodness of human nature , But selfishness makes everyone want to speed up their connection . But it's just one thing , The reason why the Internet did not collapse due to congestion , On the other hand , Not good , It's a Nash equilibrium . When there is no return on the investment , People will no longer invest , There is always a point , At the same time, everyone feels that investment will no longer have benefits .

People are selfish , Selfishness gains nothing , It becomes good .

If traditional Internet manufacturers give up transmission acceleration , Unified configuration standard BBR, Quit the competition , This is a win-win situation . WAN transmission acceleration originated from the differentiation shown by the first manufacturer , In the end, all manufacturers will follow in order to catch up with the differences .

However, differentiation does not make the screen clearer , Just a spectator in the front row stood up , Everyone stands up . When the screen is clearer , Everyone should also sit down . Such is the case ,BBR Algorithm from 2016 Years of iteration to the present , Internet transmission comparison 2014 year , Both bandwidth utilization and fairness have improved ,BBR Compared with most self-developed algorithms , It would be better, not worse .

If there is already a very clear screen , Sitting down to watch a movie is still more comfortable than standing .

Liberated workers , Can invest in IDC network optimization , This is what cloud primordial needs , Each manufacturer has centralized control , There will be no game problems . Conventional IDC The network is no longer suitable for cloud native , There are many problems to be solved , That's the point .

As I said , The fundamental of transmission acceleration is routing , Not how to contract . End to end, you can't choose the path hop by hop ,TCP/IP At the beginning of the design, I didn't want the host to control the route , trust IP Route is enough , The return of the network , The host belongs to the host .

If you have to accelerate , The correct way is to find POP Point frame SDWAN,Overlay Route selection , The essence is to choose the way , Choose a non congested road , Send it at the speed of light , Geographical distance is often not a problem .

Return to life , Traffic jam but want to go home as soon as possible , How would you choose detours and fighting traffic jams ？ Fighting traffic jams can be hard wired , Hit other cars with horsepower , You can also carefully study the reasons and details of the traffic jam , Move accurately , This experience may be useful in the next traffic jam .

Once did this , Every day is a mess , downhearted . Launch a version , Probably the observation the next day was ineffective , It may also be the normal statistical fluctuation of the network , Even if you don't do anything , Can also pretend to cheat the manager . But how to smooth the fluctuating noise , A week is too long , seize every minute , However, the amount of data is not enough , hang by a thread . It lasted for many years , During this period, some big effects seem to have been achieved only in one thought , It's hard but easy to say . This is really not a technical field , And it's really sociology . After reading the first citation paper , Conceive in the middle of the night , Looking back for a long time , Write something .

Zhejiang Wenzhou leather shoes wet , It's not fat when it's raining .