On October 20th, the Alibaba Yunqi Conference continued to be held in Hangzhou. Unlike the main forum on the first day of the opening, the main forum on the second day of the event focused more on technical fields and technical practices. On the morning of the 20th, Alibaba Cloud showed all the cutting-edge technologies such as servers, operating systems, artificial intelligence, metaverse, and data intelligence, and presented a technological feast to the participants.

The dragon lizard comes to the world, the dragon is upgraded

At the meeting, Jiang Jiangwei, vice president of Alibaba Group and head of Alibaba Cloud’s Intelligent Basic Products Division, announced that Alibaba Cloud has launched a new generation of operating system “Dragon Lizard” and made it open source. At the same time, the operating system laboratory of Alibaba Dharma Academy was also announced.

Different from the previously released Feitian operating system, the dragon lizard operating system is positioned on the server side and supports various chip architectures and computing scenarios such as X86 and ARM. It has been polished within Alibaba for 10 years, effectively supporting Tmall Double 11 over the years, and its performance and stability have withstood severe tests. At the same time, Longli has made multiple optimizations for cloud-native application development, bringing 40% comprehensive performance improvement and 50% reduction in failure rate for typical cloud scenarios. It is compatible with the CentOS ecosystem, supports one-click migration, and provides full-stack national secret capabilities. .

At the same time, Alibaba Cloud also announced the launch of the fourth-generation Shenlong architecture, which is the next-generation virtualization technology of Feitian Cloud operating system. The computing architecture innovation brought by Dragon 4.0 brings cloud computing into the era of 5 microsecond latency for the first time. Shenlong’s upgrade to io is still improving development efficiency. In the past, the supply chain system of e-commerce customers was very complicated. Once the system is not enough to support the user volume, it needs to do distributed architecture transformation. The process often requires 3 to 6. However, if it is transformed based on the Shenlong architecture, this architecture can not only support the current size of Tmall supermarket, but also support it in the next 2 to 3 years, and the upgrade does not require code modification.

RDMA is synonymous with network latency and high performance in the dedicated field. Compared with the traditional TCP protocol, RDMA can greatly reduce network communication latency. However, there are few companies on the market today that can use RDMA on a large scale, and the technology is inherently flawed. However, the current generation of Shenlong architecture overcomes this problem and realizes large-scale and inclusive RDMA networking capabilities. Alibaba Cloud adopts the design idea of ​​software and hardware integration, integrates the acceleration capabilities of elastic RMDA into the public cloud, and makes RDMA from the HPC class. applications, towards supporting general-purpose computing scenarios, and providing technical support for the explosion of cloud-native technologies such as Microservice, Serverless, and Service Mesh.

In addition, the fourth-generation Dragon also greatly improves core performance such as basic bandwidth, block storage, and IOPS. In deep learning scenarios, the fourth-generation Dragon can improve the training performance of distributed NLP and visual computing by 30%; in big data scenarios, it can improve Spark’s computing performance by 30%; in database scenarios, MySQL performance can be improved by up to 60%, The mixed read and write throughput of Redis can be increased by 130%; the throughput per second of NginxSSL connection establishment can be increased by 420%. The Shenlong 4.0 architecture has excellent performance in terms of delay. It has increased the delay from 90us to 30us, which is close to the limit of the cloud architecture and the theoretical limit of the speed of light.

Jiang Jiangwei, head of Alibaba Cloud’s basic products, said that as a representative of the new generation of virtualization technology, Shenlong was born for the cloud at the beginning of its design. A lot of optimizations have been made in large fields, bringing performance leaps for general scenarios such as databases, AI, and big data.

In terms of security, Alibaba Cloud provides enterprise-level security chips, and provides root-of-trust capabilities based on the chip, and provides full-link encryption for the running state and transmission state. At the same time, protecting customer data security is the first principle of Alibaba Cloud. , Alibaba Cloud makes its own behavior transparent, so that customers have more confidence in Alibaba Cloud.

At the same time, Alibaba Cloud supports both ECI and ECS, realizing the ultra-high elasticity of cloud native. It can support 2000 containers on a single cloud native server and support high-density deployment, which is of high value for long-tail systems, which means that users There is no need to deploy each system on a server or virtual machine, reducing resource consumption, improving enterprise efficiency and reducing costs. The Shenlong architecture is the best carrier for containers. Containers need to be isolated by nature, and both software and hardware must be used together. The AC released this year has improved the performance of open source containers by 40%. Thanks to the support of Shenlong, Alibaba Cloud can deploy 3,000 container instances in 30 seconds. It can deploy cutting-edge products such as big data at the edge, local or customer data center. , and finally achieve unified cluster management, unified asset scheduling and disaster recovery.

It is reported that the dragon lizard operating system is completely open source and provides services through open source communities and operating system manufacturers. In the future, Alibaba Cloud plans to invest 2 billion special funds for Dragon Lizard, and cooperate with 100 partners to promote ecological construction and provide technical support for at least ten years. The dragon lizard operating system has been widely used at present, and a large number of products based on the dragon lizard distribution have been deployed on Alibaba Cloud to exert their capabilities in various scenarios. In terms of ecology, through co-construction, the operating system laboratory of Alibaba Dharma Academy will cultivate 50,000 operating system talents for the society based on the community; Alibaba Cloud also hopes that the internal skills will be integrated, and through continuous opening, everyone can be based on IaaS. products, and compete fairly with Alibaba’s internal PaaS products.

At the end of the speech, Jiang Jiangwei also talked about the annual evaluation of cloud vendors’ technical capabilities by international authoritative organizations in the past three years. Through more than 200 technical evaluations, Alibaba Cloud finally got the highest score in 4 of the 9 categories. They are storage, computing, network, and security. Alibaba Cloud can achieve this achievement without the stringent requirements of customers, which enables Alibaba Cloud to make progress together with customers.

Focus on cloud native database 2.0, PolarDB three-tier decoupling heavy upgrade, PolarDB-X fully open source

With the development of industrial digitization, enterprises have higher requirements for data processing and data security, and the demand for real-time mining of data value is more and more intense. Alibaba Cloud has seen that the database has entered a new stage, and this year proposed the concept of “Cloud Native Database 2.0”. In the cloud-native database 2.0 stage, Alibaba Cloud Database hopes to move from atomic product differentiation to one-stop data management, realize one-stop full-link data management and services, and let developers, enterprise users and business side pay more attention to how Discover value from data and make it possible for data to flow freely and seamlessly.

Therefore, in the cloud-native database 2.0 era, Alibaba Cloud has created a rich portfolio of enterprise-level cloud-native database products: cloud-native relational database PolarDB, cloud-native distributed database PolarDB-X, cloud-native data warehouse AnalyticDB (ADB), One-stop Data management platform DMS, cloud-native multi-modal database Lindorm, enterprise-level cloud database service RDS and DBStack, etc.

Under the concept of cloud native database 2.0, how to carry out cloud native to the end?

At today’s meeting, Li Feifei, Vice President of Alibaba Group and General Manager of Alibaba Cloud Intelligent Database Division, announced the major upgrade of PolarDB: First, the industry’s first three-layer decoupling of memory, computing, and storage to achieve memory pooling, enabling The elasticity has been improved by orders of magnitude, and the cost has been greatly reduced; second, the multi-master architecture has been launched to further improve availability, concurrent processing, and elasticity, and efficiently cope with traffic peaks like “Double 11”; third, become a true HTAP database A system that can handle a mix of OLTP and OLAP-type workloads at the same time.

At the same time, the cloud-native distributed database PolarDB-X is fully open sourced, and is committed to building a native MySQL distributed database for government and enterprises.

Portal https://github.com/ApsaraDB/galaxysql

At the 2021 Alibaba Cloud Beijing Summit in May this year, Alibaba Cloud announced the[Alibaba Cloud Database Open Source Project]the first to open source the distributed version of PolarDB for PostgreSQL, which includes database kernels, related plug-ins, tool scripts, test cases and design documents , which is suitable for the core business scenarios of medium and large enterprises. The comprehensive open source of PolarDB-X this time is also a key step in the Alibaba Cloud database open source plan, gradually opening up the core capabilities of Alibaba Cloud database to developers and jointly promoting cloud native database 2.0.

At present, Ali is the technology company with the largest open source contribution in China. GitHub has more than 2,700 open source projects, 3,000+ contributors, and 1.11 million Stars.

Ali Lingjie debuts, an out-of-the-box big data + AI integrated platform

Since the Dartmouth Artificial Intelligence Summer Research Program in 1956, artificial intelligence has been around for 65 years, with many ups and downs. Jia Yangqing, vice president of Alibaba Group and head of the Alibaba Cloud Computing Platform Division, said that the era of AI craftsmen is bound to pass away, and we need to sort out production relations to welcome the further growth of artificial intelligence.

The key technology is the integration of big data + AI. In Jia Yangqing’s view, today’s integrated big data + AI platform must meet the following four standards:

(1) Scale: Big data, big models, and big applications;

(2) High efficiency (Speed): that is, the ultimate operation, development, operation and maintenance efficiency;

(3) Simplicity: Platform API and SDK standards, as easy to use as calling functions;

(4) Scenario: It needs to be created for the scenario and provides the platform scenario-based adaptation capability.

At present, the Alibaba Cloud big data + AI integrated platform has integrated capabilities such as data production and collection, data storage and analysis, data development and governance, and machine learning.

How to better open these big data + AI integration capabilities to users, and benefit big data + AI technology to the whole society and various industries?

In response, Alibaba Cloud officially launched the “Alibaba Lingjie” platform, allowing enterprises and developers to “out of the box” Alibaba Cloud’s big data + AI capabilities, lowering the threshold for AI applications, focusing on application development, and providing plug-in business delivery.

Alibaba Lingjie integrates Alibaba Cloud machine learning platform PAI, cloud native big data computing service MaxCompute, real-time computing Flink version, big data development and governance platform DataWorks, real-time data warehouse Hologres and other products. The high-performance training and inference engine integrating cloud, edge and terminal can provide real-time data analysis capabilities with millisecond-level delay. It is currently the largest big data + AI integrated platform in China.

In addition, in order to create a universal AI model, based on Ali Lingjie, DAMO Academy developed a multi-modal pre-training model M6, whose parameters have been upgraded from trillion to 10 trillion, making it the world’s largest pre-training model, surpassing the previous scale. The 1.6 trillion Switch Transformer model released by Google.

At the meeting, Jia Yangqing demonstrated the use of PAI and M6 with Rhino Manufacturing, using Ali Lingjie, and using simple codes to generate T-shirt production ideas in seconds. He also conducted targeted drug research with Professor Luhua’s group from the School of Chemistry and Molecular Engineering of Peking University, and used Ali Lingjie to complete the construction of the largest protein-ligand complex kinetic dataset, with a 100-fold increase in efficiency.

The ignited metaverse may be the origin of the future revolution

The metaverse is very popular recently, but it is difficult for anyone to give its true definition. So what exactly is the Metaverse? Tan Ping, head of Alibaba XR Lab, believes that the Metaverse is the entire Internet on AR/VR glasses. He believes that the metaverse is a presentation mode on the new platform, and its future may include social networking, e-commerce, education, games and even payment, etc. Today, all Internet elements will eventually appear in new presentations on the metaverse.

The metaverse can bring more than novelty, it is likely to become a new generation of technological revolution. In the 1990s, PCs were the mainstream computing platform. Later, the rise of mobile phones replaced PCs. Today, many people believe that VR/AR glasses will become the next generation platform. The iteration of the platform means the handover of technology and enterprises, the emergence of mobile phones has created a large number of local life service giants, and the new features based on the new platform will create unprecedented new experiences. In the VR/AR era, when we have virtual avatars, everything is likely to change.

The technical essence of the new platform is new Display and new interaction. In the past, whether it was a PC or a mobile phone, the Display interface was two-dimensional, and applications were built based on windows. User interaction was mouse clicks and drags in the window. Under the new VR/AR platform, display and interaction have become three-dimensional space, and we can interact with the application through body movements, language, gestures and even eyes. Display and interaction are the bottom layer of all applications. When the bottom layer changes, the upper layer application will inevitably form a revolution.

The Metaverse will bring major changes to the Internet, and new giants will likely rise again. So what levels of content are included in the technological composition of the metaverse? In Tan Ping’s view, the metaverse mainly includes the following four layers:

The first layer, holographic construction. This layer needs to build virtualized content and provide users with a basic virtual experience;

The second layer, holographic simulation. This layer will make the virtual world infinitely close to the real world based on holographic technology, and technological applications such as digital twins are also realized at this layer;

The third layer is the fusion of virtual and real. This layer needs to build a high-precision three-dimensional map based on the entire real world, realize positioning and superimpose the virtual world and the real world. If this layer is realized, it will be expected to break the boundary between the virtual world and the real world;

The fourth layer, virtual and real linkage. If there is no linkage, then beauty will only remain in the virtual world. And to transform the real world by changing the virtual world, the answer is self-evident, that is, the power of robots.

Finally, Tan Ping believes that if the technologies of these four-layer metaverses can be realized, then a new Internet era will come.

Digital inclusiveness, the trend of integrating with cloud nails

Today, there is a big contradiction in the development of the digital economy. The digital demand of enterprises is very strong but the software supply is insufficient. Ye Jun, vice president of Alibaba Group and president of Alibaba Cloud Smart Dingding, said that the reason for this problem is that the cost of software production is high, while cloud Nail one can solve this problem.

After the launch of DingTalk 6.0, the value of DingTalk is increasing day by day. It can move communication online, collaboration online, sales and other invoicing systems online, and business links of ecological partners. to the line. And what companies need is more than that. They also want data to flow, which is precisely the biggest difference between digitization and informatization. Therefore, DingTalk’s method to help everyone to enter into digitalization mainly includes two aspects: organizational digitalization and business digitalization, and these two digitalizations are called the genetic double helix structure of enterprise digitalization.

In the past, organizational digitization was a simple operation such as HR system, but now many enterprises also hope to use data to analyze whether the organizational structure is reasonable and whether the collaboration between departments is efficient. The value of organizational digital technology lies in helping management and reference. DingTalk’s fully modular integrated intelligent platform can open up massive SaaS software in the industry, make data flow, and make SaaS software connected and not isolated.

The value of business digitization lies in making business efficient and innovative. In the past, ERP data was often inaccurate and untimely, but today the problem still exists. To solve this problem, Dingding starts from four aspects:

First, put business systems on the cloud, not on-premises. In this way, applications can be moved to the cloud and innovative usage methods can be generated. Used in a unified interface and system to increase application mobility.

Second, solve the system connection problem and make the system connection more inclusive. The essence of system connection is the central problem, and the integration of cloud nails allows the problem to be solved centered on the customer, rather than centered on the platform.

Third, the problem of human-computer interaction. In the past, human-computer interaction often used components to make selection easier. But nowadays human-computer interaction still relies on a large number of input forms such as IoT devices and Internet forms, which are slow and naturally difficult. Dingding builds a low-code platform, allowing users to build applications directly on their mobile phones, providing a low-code platform and selecting 600 sets of templates to encourage everyone to embrace low-code, so that more people can develop the system and provide efficiency.

Fourth, the scene problem. The value of DingTalk is that every position on its interface can be developed, the workbench can be customized, and these capabilities are normally open to the outside world, which can help users develop across organizations and companies. Cloud Nail Integration is helping enterprises to accelerate digitalization, and it is hoped that more and more enterprises and experts will be able to enjoy digital dividends in the future.

Epilogue

As early as at the Alibaba Cloud Summit in May this year, Zhang Jianfeng, Dean of Bodhidharma Academy and President of Alibaba Cloud Intelligence, said that there are four key words for Alibaba Cloud this year: good service, solid foundation, solid platform, and good service. Strong ecology.

From the cloud chip Yitian 710, Panjiu self-developed server, and open source Xuantie RISC-V series processors released at the opening ceremony of Yunqi Conference yesterday, to today’s heavy release of Shenlong 4.0, Dragon Lizard operating system, PolarDB upgrade, PolarDB- X is fully open sourced, and new technology products such as Alibaba Lingjie are launched. We found that the strategic achievements of Alibaba Cloud’s deep foundation have surfaced, and the underlying self-developed technology has ushered in a big explosion. The ultimate goal is to create a new type of future-oriented, cloud-centric computing system.