Alibaba Cloud Yang Jingyu: General knowledge of the edge computing industry and the technological evolution of Alibaba Cloud ENS
Posted May 26, 2020 • 8 min read
Recently, Alibaba Cloud Yang Jingyu gave a keynote speech on "5G Infrastructure-Alibaba Cloud Edge Computing Technology Evolution" at the CSDN Alibaba Cloud Core Technology Competitiveness Online Summit, focusing on industry and technology trends and edge computing in the 5G era Industry knowledge and the evolution of Alibaba Cloud edge computing from the past to the future will be shared.
5G + AI demand pushes edge computing to continue to advance
Looking back on history, it has been more than 20 years since accessing the Internet in 1994. From the perspective of application scenarios, the terminal has gone from the PC era to the mobile era, including TV, camera, IoT and other terminals. The infrastructure behind it has also changed from the early self-built server to the current cloud model of the entire station. On the whole, the network architecture is a two-tier cloud and end architecture.
When 5G comes, the consumer Internet gradually extends to the industrial Internet. The Internet of Things, Industrial Internet, Internet of Vehicles, healthcare, online education, enterprise remote video conferencing and other large-bandwidth, low-latency, high-traffic scenarios are activated. The original cloud The two-layer network architecture at the end and the end is difficult to meet the needs of low latency, low cost, and localized computing under the new industrial Internet, which has contributed to the first wave of edge computing applications. At the same time, a key feature of 5G is that, unlike 2G, 3G, and 4G pipelines, 5G allows offloading at any location in the base station or network, so 5G also provides technical feasibility for edge computing. The above two points are the core reason why edge computing is gaining popularity in the 5G era.
In the cloud and terminal two-layer network architecture, computing occurs on the cloud, or on terminals such as mobile phones, PCs, and TVs. With the popularization of the IoT Internet of Things, industrial Internet, Internet of Vehicles, smart home, smart city, After the emergence of new industrial applications such as autonomous driving, cloud office, 3D video, etc., all the calculations that occur in the cloud may not meet the delay requirements, and all occur on the end side, and the requirements on the end are relatively high.
Therefore, the calculation has gradually migrated from the original cloud two-layer structure to the location above the base station and below the cloud, to achieve a balance of distance and efficiency in such a location as the city, and evolve into a three-layer structure of cloud, edge, and end. The calculated distribution pattern.
Yang Jingyu:2020 is an epoch-making year, a turning point from the consumer Internet to the industrial Internet, and a year of real success in the field of edge computing. It is expected that 5G edge computing can be commercialized and truly launched.
General knowledge of the edge computing industry:technical architecture, industrial collaboration and application scenarios
As mentioned earlier, 5G natively supports offloading in the communication pipeline for edge computing, which technically has the possibility of sinking the calculation to the base station. That is to say, before 4G, one end in the communication network and one end in the cloud. If the communication network is compared to a highway, it has no service area, and all information flow is flowing from this highway. 5G has a very good feature, which is to allow the construction of service areas on highways. When the information is at the location of the base station, it can be shunted out for local calculations. The calculation of any location of the entire communication network, such as a city.
In terms of scenarios, a large number of edge computing demands such as "low latency, high reliability/large bandwidth/large connection, low cost, and localization" will emerge in the future for immersive consumer computing and the industrial Internet. In the future, some industrial applications of 5G edge computing will emerge in large numbers. At the same time, the cloud computing of the terminal, such as cloud gaming, cloud desktop and other application scenarios, is the best state in terms of the cost to users, and the cost and maintenance of enterprises and equipment.
There are many players in the upstream and downstream of the edge computing industry, mainly including operators, cloud vendors, professional service vendors, and equipment and chip vendors, etc., which provide assistance for the development of the edge computing industry from their respective advantages.
Cloud vendors have always provided a pay-as-you-go cloud service. With the advent of the Industrial Internet, cloud vendors will further extend and expand cloud services in conjunction with edge computing. Large-scale and centralized computing is still left to the cloud center. , Cost, and localized computing have strong demand, they are placed on the edge to complete. For operators, edge computing is calculated at the operator's metropolitan area network or 5G core network, base station and other locations. The calculation takes place in the operator's entire network system, whether it is the three major foreign or domestic operators. The current investment in edge computing is very large. Professional service vendors include some CDN vendors. The calculations he originally defined are actually placed in prefecture-level IDCs. They are not strictly the edge of the communications Internet, but also a kind of edge. In addition, part of the edge computing is equipment, such as edge computing in the field of gateways, which involves the field of equipment manufacturers and chip manufacturers.
At present, the application scenarios of edge computing are roughly divided into two categories. The first category is to deploy such distributed computing resources across the country or even around the world. The typical application is the CDN content distribution network, which is actually provided on distributed nodes. Content caching and forwarding. Future online education, online meetings and other scenarios are also core scenarios of distributed Internet applications. The other category is industrial applications, which require localization and low latency for computing, which may be completed within 10 kilometers and 100 kilometers, such as video aggregation, smart parks, and other scenarios.
Edge node service:open, standard, flexible and easy-to-use edge cloud operating system
Talking about the difference between public cloud and edge cloud and the main technical characteristics and challenges of edge cloud, Yang Jingyu believes there are four main points.
First, both public and edge clouds are large-scale, but the scale attributes are different. The public cloud is a centralized large-scale cloud, usually composed of a small amount of large resources, and edge computing is usually composed of a large number of small nodes. At the same time, from a technical point of view, the entire security system, computer room, power, network environment and edge cloud environment of the public cloud are not the same, so the construction and asset management logic of the edge cloud and the public cloud are different. This is also a very big challenge.
Second, the edge cloud is not a cloud for every small node. Yang Jingyu believes that the edge cloud is a cloud composed of tens of thousands of nodes or millions of nodes. Therefore, the edge cloud needs centralized management and unified scheduling based on the formation of an unreliable Internet. Even when the edge node has a network interruption, it can be off-network autonomous.
Third, edge computing must undertake various network connection requirements of the Internet of Things and the Internet of Vehicles. Therefore, edge computing must solve the problem of how to do 5G network offload, including what kind of traffic needs to be offloaded. How to record the stream, it is not simply to provide a computing service, but to provide cloud and network to provide customers with the use.
Fourth, in terms of how customers use it, in the past, the common way everyone used to use public clouds was to buy a region, several machines, and an IP, and they would know where it is. But edge computing may actually be a disaster if it follows this form.
Yang Jingyu made an appropriate analogy:In fact, using edge computing should be like we are using a phone. When using the phone, I will definitely use a 4G or a base station. Even when driving, the mobile phone will continue to connect to the base station. There has been a change, but as a user, I don't care where the base station is. I just enjoyed the service of making a call. I hope that the service is innocent to me. Therefore, for this large-scale edge computing infrastructure, it is not simply to sell the computing power of users one node at a time, but to consider how to provide users with a more seamless experience and great help in cost. Such a computing system. We have always compared ENS to an operating system. The characteristic of the operating system is that my application uses CPU and memory based on processes and threads. Users do not need to care which core of the CPU and which time slice it uses. Calculation. Therefore, all the ENS operating system needs to do is manage the computing power and use it seamlessly, which is a more difficult part to handle than the public cloud.
As shown on the right side of the above figure, the bottom layer of Alibaba Cloud ENS is still a variety of infrastructures. Based on this infrastructure and the virtualization of Alibaba Cloud and the technical precipitation in network computing storage security, a large-scale A distributed cloud operating system or edge computing platform that is unknown, unknown, and insensitive, shielding the underlying resource differences, scale differences, and usage differences. Through the opening of capabilities and linkage of operation and maintenance, it provides a more convenient infrastructure for upper-layer applications.
Alibaba Cloud has been technically polished since 2017, and it has been transformed into an edge computing node based on CDN nodes all over the world. After 2 years of development, in 2019, Alibaba Cloud edge computing has actually achieved more than 300 locations in China. The global coverage of the city is controlled within 10 milliseconds. That is, when the user goes to watch the video, 25 frames of video may need to be played for 40 milliseconds, then 10 milliseconds will make the user's vision without frame perception. For cloud games, cloud desktops and other applications, it can be obtained on the delay Great satisfaction.
The concept of edge computing has been very hot in recent years, but there is still a certain gap in understanding the definition, including the implemented technology stack. For the positive and benign development of the edge computing industry, Alibaba Cloud has also invested a lot of energy in the industry standardization. In December 2018, Alibaba Cloud took the lead in releasing the edge cloud white paper, which defined the concept of edge cloud computing for the first time in the industry. At the Yunqi Conference in September 2019, Alibaba Cloud and more than 20 units released the general technical requirements standards for edge cloud computing. At the same time, in December, Alibaba Cloud also won the first award issued by the China Electronics Standardization Institute The approval of the "Edge Cloud Standard Conformance Test Certificate" also means that Alibaba Cloud is taking a step forward in standardization in the field of edge cloud.
Yang Jingyu believes that Alibaba Cloud Edge Computing is to build urban computing, that is, to gradually deploy edge cloud infrastructure in units of prefecture-level cities, and to help Alibaba economies and industry customers implement various applications to the edge cloud. Life in a smart city to empower all things in the 5G era will come sooner.