Netease Cloud Chief Architect: How to implement solutions and productization practices

Our architecture is mainly divided into three aspects: business architecture, IT architecture and data architecture, which solve the problems of users, operation and operation respectively.

For IT architecture, we provide basic computing, network, storage and other resources through basic services. On the business architecture side, it is possible to deploy a fully elastic architecture based on basic services, which has many best practices, such as Koala, cloud music, cloud classroom, etc. If you want to do live streaming, you can access the communication and video cloud in just a day. For data architecture, we have two products, one is Mammoth big data, the other is BI. Artificial intelligence has seven fish and easy shield two products.

Through the service of a large number of traditional and Internet-based products, we found that data is a very important aspect in both e-commerce and games, and the core of different business models in the future may lie in the difference of data. For example, a traditional automobile retail enterprise may need an operation team to collect scattered data and issue a report every two weeks to guide the next step of operation, while an Internet-based e-commerce double Eleven needs to see real-time operation data every second.

This is a more detailed architecture, the architecture on the right is the infrastructure layer, and the architecture on the left is mammoth Big Data, which has security, digital, artificial intelligence, video cloud, IM, etc.

The infrastructure layer of our cloud computing infrastructure platform is different from other applications because we think containers are the way of the future. Container is a container of software supply chain, which can realize cross-environment migration based on image. One aspect of implementing cross-environment migration is to migrate between development, test, and production environments for continuous integration. In many of our internal practices, the biggest benefit of using containers is that the container allows the entire delivery process to advance to development, delivering a mirror image of the container that contains most of the configuration and allows for rapid iteration.

Another aspect of implementing cross-environment migration is flexible capacity expansion. The right part of the figure above is a de-statification process that externalizes everything local, allowing for horizontal elastic scaling if the application contains only one program logic.

Containers can also help build big data platforms. There are two types of big data customers who migrate to netease Cloud. One is that they have big data capabilities and purchase or customize big data by themselves. The other is using our mammoth big data. For the first type of users, the biggest problem with big data containers is state. It would be complicated if HDFS were stored in containers, but now that the network is not a problem, data can be stored in independent HDFS clusters or object storage, so Map Reduce becomes a stateless process. The biggest advantage of this is that you don’t have to maintain such a large cluster. This makes it possible for containers to deploy big data in the future.

Another typical use of containers is microservices. When a service takes on a large volume, it needs to be split. However, if you split the backend, a single database will not be able to support it. DDB distributed database is usually used to do this. All the service scenarios in the middle are stateless and can be flexibly scaled.

Proprietary cloud ensures IT architecture security

Today’s topic is big data, we found that everyone is very concerned about the security of their data, so we launched a new concept, exclusive cloud. People often confuse private clouds with VPCS, private clouds, hosted clouds, and proprietary hosts. A VPC is logically isolated, while a private cloud is physically isolated. Private clouds are also different from private clouds. Private clouds are deployed in the customer’s own computer room, and the operation, maintenance, and upgrade of private clouds are the customer’s responsibility, no matter the deployment of big data or cloud computing platform. Exclusive cloud and hosted cloud is not the same, managed cloud operation and maintenance, upgrade is also the customer to do, just do not need to build their own room. The exclusive cloud is deployed in netease’s data center, and all its operation and maintenance, upgrading and upgrading are done by netease. In this way, there is no need for many operation and maintenance personnel to manage the cloud platform, which can be handed over to netease Cloud. Exclusive cloud can ensure that all computing and storage resources of users are stored on their own machines. When a single user buys resources, the hardware and software will be automatically upgraded and the latest functions will always be used. Many cloud platforms also launched their own host, cloud host can be their own. But our proprietary cloud is from the bottom up, from IaaS, to PaaS, to CaaS.

Big data and BI realize digital operation

With this platform, the next thing to do is how to use this platform to really do operational capabilities. Big data can be analyzed in three ways. The first is transaction data. After customers place orders, all data are stored in the database, which is a very important source of big data analysis. If you use a proprietary cloud, it will be your own.

The second part is the log data of users’ browsing and clicking. Log is generally done in the way of ELK in microservice architecture. Since microservices have been deployed, there is a centralized place to read logs, so there is usually a log library. When encountering a big push, you need to degrade some parts and asynchronize others.

The third part that is most overlooked is customer service data. The performance of customer service during the promotion period can also be counted afterwards, which can be put into the big data platform and finally put into the data warehouse for the BI platform to see.

Meow is a very good digital operation platform. The general analysis will have two dimensions, one is the vertical dimension, is the time axis; The other is the horizontal dimension, which is layout. There will also be some analysis models, such as the people store model, user browsing model, etc., which can be analyzed.

Based on artificial intelligence technology, netease Cloud has eshield security plan and Qiyu customer service plan.

Yi Shield can monitor user login and cheating behaviors, protect user login and registration through ARTIFICIAL intelligence, and take corresponding measures against cheating behaviors.

If there are some bad videos or pictures, image recognition can be detected. It used to be based on probabilistic models, but now it’s going to be based on deep learning, using neural networks.

Seven Fish intelligent customer service at first through keyword matching, then probabilistic model NLP, now through deep learning and neural network gradually developed to today.

This is the overall solution from the bottom to the top, you can consider these three aspects if you go to the cloud.

That’s all for today’s sharing, thank you!