Notes on "Alibaba China-Taiwan Strategic Thinking and Framework Actual Combat"

 

Recently I was reading a book called "The Transformation of Enterprise IT Architecture: Alibaba’s Strategic Thinking and Architecture Practices in China and Taiwan". At the time of writing this article, the book has not been read, because I am worried that if I read all the books, I will write this. This article, a lot of wonderful content may have been forgotten, so I will write an article to share with you.

 

China-Taiwan Strategy

 

Notes on "Alibaba China-Taiwan Strategic Thinking and Framework Actual Combat"

 

The Taobao Affairs Department established by Alibaba in 2003 is shown in Figure 1.

 

In 2008, the B2C business was booming, and Alibaba established Tmall, initially called Taobao Mall, which was then operated as a division of Taobao Business Unit, as shown in Figure 2.

 

With the continuous increase of B2C business, Tmall began to become independent. Alibaba established the Tmall business department separately, alongside Taobao's affairs department, as shown in Figure 3. At this time, Taobao's technology part supports both business departments at the same time. This organizational structure It is determined that the technical team will give priority to meeting the business needs of Taobao, which has seriously affected the development of Tmall's business. Anyone who has used Tmall and Taobao should be able to find that e-commerce platforms such as Tmall and Taobao all include functions such as commodities, transactions, evaluation, payment, and logistics.

 

In 2009, the shared business division came into being. The main members came to the Taobao technical team, and in terms of organizational structure, it became a division of the same level as Taobao and Tmall, as shown in Figure 4. The Group hopes that in this way, the technical team can support both Tmall and Taobao businesses, while at the same time accumulating public and general businesses, and making more rational use of resources.

 

But in fact, at the time, the shared business division was "submitted to" Tmall and Taobao. The shared business division needed to meet the large needs of Tmall and Taobao at the same time. Team members often worked overtime and might not be able to meet the requirements of Tmall and Taobao. Demand, this has caused the business departments of Tmall and Taobao to be unsatisfied with the shared business department, and at the same time, the colleagues of the shared business department can only have troubles.

 

In 2010, the group buying business Juhuasuan appeared. Juhuasuan has strong traffic attracting capabilities. Therefore, Tmall, Taobao, and 1688 all want to connect with the Juhuasuan platform to expand their traffic. Suddenly, Juhuasuan is overwhelmed by such a large docking demand. At this time, the group requires that if the three major e-commerce platforms want to connect to the Juhuasuan platform, they must go through the shared business division! It is with this policy that the shared business division has a strong business grasp, and has brought the original imbalance with the three major e-commerce platforms to a relatively fair level. This has established that the shared business division that everyone sees today has become the core business platform in the business of Alibaba Group, as shown in the following figure:

Notes on "Alibaba China-Taiwan Strategic Thinking and Framework Actual Combat"

 

The above figure clearly describes the structure of Alibaba's "thick platform, thin application", and the shared business division is the true embodiment of the "thick platform". The "thick platform" provides the most professional and Stable business services, this is Zhongtai .

 

We can find that the China-Taiwan strategy does not happen overnight. When the shared business division was established in 2009, it had already laid a certain foundation for the China-Taiwan strategy. At the same time, it also needed the strong support of the group to build the China-Taiwan. Formation has laid a solid foundation for the development of business.

 

Chimney architecture

 

In 2008, Taobao’s technical team supported both Taobao and Tmall’s two major e-commerce platforms. At the same time, 1688 had its own technical team. The structure is as follows:

Notes on "Alibaba China-Taiwan Strategic Thinking and Framework Actual Combat"

 

This architecture is a chimney architecture . Each business department and their corresponding business department stand there like a chimney, and if according to this architecture, when the enterprise needs to expand new business, a new business department and corresponding The new technical department has an extra chimney.

 

So far, this kind of architecture still has many companies like this. The reason for this kind of architecture must have its advantages:

  • Companies consider different business models, so they build independently
  • The new business team believes that the transformation on the basis of the previous business will have too much technical and business historical burden, it is better to rebuild

 

It's just that the disadvantages of this architecture far outweigh its advantages:

  • Repetitive function construction and maintenance bring repetitive work and investment. Repeated construction can reduce risks for enterprises, but it will increase repeated costs.
  • If the "chimney" system interacts, the cost of collaboration is high.
  • It is not conducive to the precipitation and sustainable development of the business. After a chimney goes online, it enters the operation and maintenance stage. At this time, if you need to modify the business on this basis, it will take a long time to publish the business.

 

In the Internet age, it is inevitable to better integrate enterprise internal resources, reduce enterprise costs, and realize the interaction between various systems. Faced with this situation, in 2004, the industry has put forward the SOA concept to solve the problem of interaction between "chimney" systems.

 

SOA

The core functions of SOA:

  • Service-oriented distributed computing
  • Loose coupling between services
  • Encapsulation of Support Services
  • Service registration and automatic discovery
  • Define service interaction methods in the form of service contracts

 

Centralized SOA

Many enterprises implement SOA through ESB, which is a centralized SOA.

 

ESB is an enterprise service bus. As the name implies, the ESB system can uniformly manage various services in the enterprise. The ESB architecture shields the impact of service interface changes on service consumers, and is to solve the problem of different systems. A very good architecture for interconnection, as shown in the following figure:

Notes on "Alibaba China-Taiwan Strategic Thinking and Framework Actual Combat"

 

In 2004, many large software companies have discovered that more and more companies have gradually built more and more IT systems in the course of years of IT construction. These IT systems are all established by using the chimney system construction model. There are many systems in the enterprise. Some of these systems are purchased commercial kits, some are independently developed, and some are developed by outsourcing companies. The final result is that the technology platforms, frameworks, and languages ​​used by each system are different. So software companies have developed ESB systems to help these companies solve these problems.

 

The service provider only needs to define the interface and the access path of the interface on the ESB system. It does not need to care about who is the consumer of this service, and the modification of this service only needs to be adjusted once in the ESB. It realizes the isolation that affects the service interface changes. ESB reduces the coupling between systems, realizes system integration more conveniently and efficiently, and provides more professional capabilities than the "point-to-point" model in terms of service load balancing and service management and control.

 

ESB provides functions such as adaptation and access to various technical interfaces (HTTP, Socket, JMS, JDBC, etc.), data format conversion, data tailoring, service request routing, etc. The purpose is to allow enterprise customers to improve development efficiency based on these functions , Realize the project landing faster.

 

Therefore, the ESB approach has become the mainstream of SOA implementation in this period, and it has solved the interaction between heterogeneous systems.

 

Decentralized SOA

 

"Decentralized SOA" is brought by the Internet industry, because the user group in the Internet industry is the entire Internet public, so system architects must first solve the problem of system scalability in order to respond to business faster , Better support for business innovation, etc.

 

Therefore, in addition to satisfying the core functions of SOA, "decentralization" must also avoid the difficulty of scalability caused by "centralization" and the potential "avalanche" impact.

 

"Decentralized SOA" is a "point-to-point" architecture, it has no center, as shown below:

Notes on "Alibaba China-Taiwan Strategic Thinking and Framework Actual Combat"

 

Then there may be doubts. The emergence of SOA is to solve the problems caused by the chimney architecture, and the calls between the chimney systems are "point-to-point". Isn't this going backwards? In the Internet industry, the decentralized service framework runs inside the enterprise, and service interactions across intranets and extranets are rarely seen. In addition, the service is contract-first, and the service interface functions are agreed upon, which is good to some extent. The stability of the service interface is guaranteed. At the same time, the decentralized service framework plus the support for functions such as multi-version and load balancing essentially shields various system instability problems in the previous "point-to-point" mode.

 

In the "centralized architecture", the center of the entire architecture is the ESB, and all service calls and returns must pass through the ESB, so that the service caller will have a lot of network overhead when calling a service, while in the "decentralized architecture" ", this problem will not arise.

 

In addition, all service calls go through the ESB, so ESB cluster deployment is inevitable. In addition, in order to ensure that there will be no problems with the ESB, the server configuration or network configuration of the ESB system will be better deployed, which makes it necessary for enterprises to expand the ESB , Will bring about a significant increase in software and hardware costs.

 

In addition, even if the ESB system uses cluster deployment to ensure high availability, there may still be an "avalanche" effect. Once an "avalanche" occurs, all services in the enterprise will be unavailable .

 

avalanche

We assume that the maximum concurrency of each server in the ESB cluster is 100. Assuming that there are 10 servers in the cluster now, when the daily user request volume is stable, the average concurrency of each server after load balancing is 80, but if the cluster One of the servers suddenly fails. At this time, another 9 servers are needed to bear the previous concurrency. Then the concurrency of the remaining 9 servers will increase, which may cause one of the 9 servers to be overwhelmed. , Resulting in the remaining 8 servers being crushed one after another, this is the "avalanche." Once an "avalanche" failure occurs, even if you restart the server, it is difficult to solve, because it is very likely that the server will be overwhelmed by the traffic just after the startup is completed, so at this time you can only prohibit external traffic from flowing into your system. Wait for all servers to start successfully before letting in traffic. And when this happens, you may not have time to locate the problem. The restarted cluster is actually in a "fragile" state.

 

This means that the "centralized" architecture cannot solve the problem of system scalability, and the "decentralized" architecture will be better, because even if the above situation occurs, it will not affect all services. So this is why the Internet industry chooses a "decentralized" architecture.

 

Below we introduce the Alibaba Distributed Service Framework HSF , wait until I finish reading it...haha.

Guess you like

Origin blog.csdn.net/yueyunyin/article/details/109162077