How did the data center change from "Britney Spears" to "Mrs. bull"?

6 month 29 The latest news from Japan , Alibaba Group merges the original data to Zhongtai 、 Business Center 、 Customer service system 、 Supply chain services and other departments , Create a new brand of enterprise digital intelligence services “ antelope ”—— a DAAS（ Data intelligence as a service ） manufacturer .

This news makes many insiders have questions ： This move to establish a new king , Is it a hint that Alibaba is about to put the data into the console shutdown.

Review the fate of the data center , The typical high opening and low walking makes people sigh ：

2015 year , Proposed by Ali “ Tai Chung Tai ” Data center strategy ;

2019 year , The first year of Zhongtai in historical data , Major factories and middle office service providers “ Daxing ” Data center ;

2021 year , The major factories began to dismantle the middle stage one after another .

Why just 6 Year time , How can the data center start from “ Britney Spears “ Turned into “ Mrs. Niu ”？ Is the data center really good for nothing ？ Why does the data center “ Defeat Maicheng ”？ In this article, the author will elaborate from these two aspects .

One 、 Is the data center really good for nothing ？

First , Here I must praise the concept of Zhongtai . Technically speaking , Add a middle desk between the front desk and the back desk , The design of this architecture is very reasonable . The author summarizes several reasons as follows ：

reason 1： The middle stage can shield the data storage in the background , It can also cope with the diverse needs of the front desk . If all the requests from the front office are handed over to the back office , The backstage is responsible for too many and too many things, which leads to low efficiency .

reason 2： To a certain extent, the foreground and background are two different optimization objectives , The same team deals with both platforms on the same set of hardware , Easily lead to schizophrenia .

reason 3： Multiple foreground share the same background , If the background directly provides flexible data services to the front desk , This may lead to a high degree of coupling between the foreground , Maintenance costs may double .

reason 4： It's not appropriate to put data processing on the front desk , The first is insecurity , Secondly, the application experience of the front office team , For example, better interface and smoother use , I don't have much time to think about data .

The author thinks , The middle stage allows the background to focus on data storage , The front desk focuses on the application experience , The middle desk is responsible for smoothing the gap between the front desk and the back desk . This design , It seems that the division of labor is clear and each performs his own duties , Efficiency can naturally be improved .

Two 、 Why does the data center “ Defeat Maicheng ”？

Since the structure of the middle office is so reasonable , Why can't the industry continue ？

Because the data center is just a concept and methodology , But it is not very good at the specific landing level .

How the insiders analyze it , But the author thinks that many views have not touched on the point . Because there is no problem with the concept and methodology in the data , The key issue is the landing . here , The author from the coding Make an analysis from the angle .

In a word ： The industry has no technology ready to let the data platform land ！

First of all, make sure that , What is the core value of the data center ？ Provide data services to the front desk flexibly and quickly , That is to return some appropriate data from the background to the foreground after receiving the foreground request .

The next step is to see how to return data ？ The ideal approach is to calculate , It is to move the work done by the database in the background to the middle stage . Then , What techniques are used to write these calculation codes ？ Now I will analyze it one by one for you .

choice 1：Java

Are you kidding ？ Write a slightly more complex group summary, which can be hundreds of lines , How to improve efficiency ？ Also want to quickly respond to changes in the front desk ？ It takes days to write and tune the code . The task of the middle office , It is also a task before the database , Most of them are structured data related computing . and Java Such a high-level language has no easy-to-use structured data computing class library . Originally used SQL Something that can be solved , Now use Java It takes hundreds or even thousands of lines of code . The code is too long , It is not only difficult to write but also easy to make mistakes . and ,Java The cost of programmers is high . This may lead to , The efficiency of the data center has not been improved , The cost is getting higher and higher .

choice 2：Stream

Someone might say ,Java Support Stream These problems will be solved in the future . however ,Stream It's good to watch , Actually, it is not the case at all .Stream The intermediate calculation result and the final result must be defined in advance , The definition and assignment of structure are very troublesome . If not defined , Reading and using are not intuitive .

and Stream Although support Lambda grammar , But the interface rules are complicated , Code is not much shorter, but dyslexia has increased significantly .Stream Structured objects such as record\entiry\Map It's not convenient , The root cause is still that Java Lack of professional structured data objects , Lack of strong support from the bottom .

choice 3：Kotlin

And Stream similar ,Kotlin Insufficient computing power , It also lacks professional structured data objects , Cannot support dynamic data structures 、 It's hard to really simplify Lambda grammar 、 Fields, etc. cannot be referenced directly . meanwhile ,Kotlin It also lacks some important basic functions , For example, correlation calculation , Developers still have to hard code to complete the calculation , For the business algorithm composed of multiple basic calculations , The development process is still difficult .

however , The middle office structure of some large factories has been well implemented , How can this be explained ？ The author thinks , Maybe it's because there are many talents in big factories ,Java Accumulate a lot of code , It's a little easier to implement these calculations . however , Although the Internet giant is large in scale , However, the business complexity is far behind that of traditional enterprises . therefore , Internet companies can run smoothly , Traditional enterprises may not be able to run smoothly . What's more? , Big Internet companies have begun to dismantle the middle stage ？

choice 4：SQL

Java、Stream and Kotlin Will do , use SQL No row ？ Sure , But we have to put a database in the middle stage , Migrate a pile of data from the background to the middle stage . How much data to migrate ？ It seems that all the data can be used to calculate , Then we have to move the whole background data .

However , Can this still be called the middle stage ？ I just moved the backstage . There is no database independent 、 Can be integrated and embedded 、 Support heterogeneous data sources 、 Simple, convenient and powerful structured data computing power , The data center can only talk about fantasy , The architecture is good-looking but can not be implemented .

Some people say , Can you force yourself onto the middle stage ？ Unless the business is simple enough , Otherwise, it will only increase the development cost and reduce the efficiency —— Flexibility has not increased at all , There's a lot of trouble .

Taken together , The data center must be behind the computing engine with the above characteristics , Only in this way can the reasonable structure of the data center really play a role , Only in this way can the data center be truly grounded 、 Bloom 、 result .