Introduction: Since the first year of 5G commercial use in 2020, all kinds of edge scenes have begun to heat up, and edge computing has returned to people’s vision. This return is accompanied by the popularity of cloud computing and the subversive development of communication technology. As the relay node of 5G and central cloud computing, edge cloud is in the key position of cloud network integration and connecting the preceding and the following. And edge cloud + cloud native will be like container changing the world, based on the standardization and profound accumulation of cloud native, better solve all kinds of problems of edge heterogeneous, vertical, distributed and other scenes enabling edge scenes. Based on this thinking, ali Cloud has gradually built the original edge cloud system in combination with the rich edge ecological business of Alibaba Group. In the recent CCF TF48 seminar, Zhou Jing, senior technical expert of Ali Cloud Edge cloud, shared the thinking and precipitation in the construction process of Ali Cloud edge cloud primary system for on-site technical experts.
Zhou Jing ali Cloud edge computing senior technical expert
Vanilla author of high-performance OpenResty development framework, WeiboMesh founder and initial core member, promoted and completed the transformation of microblog core service Mesh, open source lover, currently mainly responsible for the design and development of Ali Cloud edge fusion computing platform and the construction of edge cloud native system. Committed to 5G edge computing infrastructure design and construction and industry landing.
Here’s an excerpt from the post:
Edge cloud development trend
Review of the evolution process of the whole mobile communication, from the 1980 s arrival 1 g mobile phone was born, and the simulation to the later 2 g network of digital mobile phone, 3 g, the worldwide compatibility and web page generation, 4 g mobile broadband speed video transmission, to 5 g era of mobile Internet of things with vertical industry depth fusion, 6 g depth at all to the future outlook, The main body of communication gradually shifted from human-centered to material-centered. The flow of information, the way data is produced and consumed, is also changing dramatically. In the 5G environment, the access capacity of wireless network is further improved, and the production and consumption mode of data is transformed from centralized production and decentralized consumption to decentralized production and ubiquitous consumption, which means that application reconstruction and industrial collaboration are required in terms of technology.
In 5G, 80 percent of data and computing will take place on the edge
With the enhancement of mobile broadband by 5G network, mass terminal interconnection and demand for high reliability and low delay connection, the value of edge cloud will become increasingly prominent. Edge cloud implements localized processing and distribution of 5G heavy traffic by converging traffic at the edge, avoiding the impact of massive traffic on the backbone network and effectively reducing the remote transmission cost of massive traffic. Relying on distributed architecture, edge cloud computing realizes distributed processing of massive terminals with high concurrency and localized computing capability of in-depth scenarios, effectively improving computing efficiency. Meanwhile, edge cloud can be deployed nearby to meet the scenario requirements of 5G low-latency processing.
Ali cloud edge cloud | a cloud core, a cloud polymorphism
At the Ali Cloud Summit with the theme of “Innovation on the Cloud” in May this year, Ali Cloud released the integrated Feitian architecture of one cloud multi-core and one cloud multi-state, aiming to provide a set of x86, ARM, RISC-V and other chip architectures with the comprehensive compatibility of Feitian operating system: Feitian elastic architecture with unified cloud product architecture and consistent experience radiates from the center to the edge, making computing power everywhere.
The four types of cloud are different in location distribution, machine room scale, coping scenarios and problems to be solved. The central Region is located in the core area of the frontline. As a large-volume and large-scale public cloud of the whole product, the central Region applies to various flexible, high-density, large concurrency, and high availability scenarios, such as familiar Internet computing scenarios, big data, AI model training, and reasoning scenarios. The central Region is far from the terminal user, and the delay is less than 100 ms.
IoT field computing node is located in the user machine room and business site, nearest to the user, to provide hardware and software integrated computing solutions, for equipment manufacturers to provide cloud-integrated intelligent equipment services, for industry and government to provide enterprise digital IoT infrastructure. In response to intelligent IoT devices, construction of IoT platforms in vertical industries, distribution and expansion of IoT devices and other scenarios, the delay is within 5 ms. The time delay of the local Region and the edge cloud node between the center and the site as relay nodes is between 5 ms and 20 ms. The local Region is located in the active digital economy Region and provides larger computing services than the edge cloud node through the miniaturization of the central cloud output. Focus on supporting regional enterprise digital transformation scenarios.
The edge cloud that our team is responsible for is composed of large-scale geographically dispersed edge heterogeneous multi-cloud fusion nodes with a scale of hundreds of orders of magnitude. The nodes extensively cover hot spots closer to users and form a distributed cloud in collaboration with each other. To cope with edge device management, cloud on intelligent terminals, view streaming, rendering, CDN, terminal computing power moving up, cloud computing power sinking, 5G + edge cloud network integration and other edge scenarios, provide users with closer, lower latency and consistent cloud services with central experience. Ali Cloud Edge cloud has the characteristics of remote control, security and reliability, and easy to use standards.
The two problem domains of edge cloud and edge cloud native
With the in-depth development of cloud computing, enterprises and organizations can focus more on their own business. A large number of central cloud demands not only promote the standardization of cloud system, the growing maturity of cloud industry and the continuous rise of relevant technology level, but also educate and guide users to have a better habit and willingness to use cloud. Based on the principle of container standardization changing the world, the concept of cloud native gradually becomes popular. It can be predicted that cloud native will inevitably play a pivotal standardization advantage in the edge in the future, because no organization will not like the optimization of cost and efficiency. The cloud native methodology is the perfect solution to the organization and full cloud of edge distributed cloud services. In the process of the long search, ali edge cloud system construction process began.
Two big problem domain, fusion edge VS | center cloud cloud
Then how to build edge cloud native technology system? What problems need to be solved in the process? What are the challenges? For cloud vendors, they certainly hope that customers will use cloud based on cloud native methods, especially in edge scenarios, due to environmental and resource factors, it brings great challenges to control and cloud services. On the other hand, from the input of scene requirements in various vertical fields on the edge, the scenes and requirements we have to face are extremely complex and difficult. For an example of heterogeneous resource integration, intelligent camera, view the device may need to use the GPU, and the other scenario is need general CPU, may also need to have storage support, then, how to face tens to hundreds of the edge of the IDC to provide service above fusion of the work force, how to minimize various controls the pool, Selling more reliable computing power is very challenging.
The fusion here is mainly the ultra-converged edge basic IaaS service and the service fusion of both sides of the cloud network. The following core problems should be solved:
- How to build edge infrastructure through hyper-converged IaaS technology;
- How to manage the distributed and heterogeneous resources in the cloud through a unified management and control system, and provide the Infrastructure PaaS (IPaaS) capability that is insensitive to the whole network and the whole domain based on the cloud native mode;
- How to reuse the current cloud native technology architecture to provide APaaS capabilities for efficient and reliable service management and deployment based on the previous IPaaS;
- How to provide users with more convenient and reliable services by integrating services on both sides of the cloud network;
- The coordination here includes communication and control coordination, such as cross-IDC coordination, and cloud-network capability coordination, such as network service coordination and cloud network resource coordination.
The following figure shows the difference in resource attributes and node attributes between the central cloud and the edge cloud:
Edge computer rooms are often relatively small in scale, ranging from a few to hundreds of machines, and widely distributed with natural distributed cloud architecture. Computing power resources are heterogeneous, and usually have specific hardware resource requirements for different scenarios. The total amount of edge IDC resources is limited, and many cloud manufacturers tend to trade off each other, so the docking of multi-cloud resources is also the only way. Network environments vary, including single-line and multi-line networks, classic networks, VPCS, and small operators whose network architectures are often fully customized. For example, computing and storage often require mixed pools. Even if an independent storage resource pool exists, it also needs to provide multiple forms of storage capabilities, such as block storage, object storage, and NAS storage. So building hyper-converged IaaS infrastructure at the edge is a natural choice.
Two big problem domain fusion | computing, storage, network and scheduling
In terms of the fusion of computing surface, we realized the service of multiple computing force forms such as bare metal, virtual machine and security container on the same logistics equipment based on the method of Cloudlet fusion computing.
Edge of computing products and research and development team, most of the CDN industry experience, have strong ability of caching technology, but for edge calculation, in addition to the services industry in the application of the Internet and the Internet, you need to provide low cost based on the Internet node, distributed storage and buffer system, multi-type, at the same time, the safety of the stored data are to be of special consideration.
When edge nodes can provide more computing capacity, how to make it easier for users to use edge computing services? Edge network capability is an indispensable part. We need to build a cloud-edge-end collaborative edge computing network to support multi-cloud collaboration (including collaboration with operator MEC network). To provide users with secure, reliable, accelerated, low-cost network services in a transparent manner.
The number of edge nodes is large and the distribution is wide. How to achieve the fast distribution of computing power and the second distribution of node image in the whole network?
Two big problem domain, fusion | cloud network integration It’s cloudy fusion MEC and 5 g network docking
5G networks are designed with edge computing support in mind, such as MEC, flexible deployment of user plane, control plane separation and user plane, and multiple local streaming technologies. It is also mentioned at the beginning that the establishment of independent edge computing related protocol specifications in 5G R17 determines the general direction of cloud network convergence from the standard. The foreseeable future with the cloud network must be you have me, I have you.
At present, with the promotion of 5G construction, operators have gradually developed relevant 5G open capacity NEF according to 5G standards. The edge cloud side is also trying to connect with operators’ MEC resources. Operators are also starting to use MEC as part of their edge cloud services to provide computing services to b-end customers. In fact, it is a win-win situation. Through the integration with cloud, operators can revitalize network resources that used to rely on relatively rigid hardware based on the characteristics of cloud computing such as intensive, elastic, on-demand expansion, and maximize their own resource advantages and network capabilities and values based on cloud network integration. Cloud vendors interconnect with carrier resources and services to provide customers with edge cloud services with lower latency and higher security.
Two big problem domain · | network synergy together
The most basic distributed collaboration of edge cloud native comes from the natural distributed environment. Each edge IDC is an edge Cloudlet unit. In the process of management and control, multiple adjacent independent Cloudlets can be integrated into a larger scale unified Cloudlet management. Mainly deal with Pod migration across IDC; The edge single node is off the network, how to return to the cloud through the neighboring node or cooperate with other nodes communication; Ali Cloud edge cloud has built a relatively rich network collaboration capacity, and comprehensively responds to various problems and challenges brought by network collaboration from end-to-end, edge-edge and edge-cloud aspects.
From the perspective of the distribution of the original ecosystem of the edge cloud, we hope that the control of the original ecosystem of the edge cloud is gradually decentralized, although the current practice is more centered control and edge autonomy. However, from the perspective of scale, long-term development will inevitably lead to the huge scale of the edge, which can easily overwhelm the centralized management and control mode. Therefore, it is wise to consider the decentralized distributed cloud management and control mode at the initial stage. The decentralized and unitary distributed cloud management and control will put forward higher requirements for distributed collaboration. In the near future, edge grid and distributed message bus will be the core technologies to solve these problems.
Edge cloud native | edge cloud network integration
As more and more edge computing nodes carry more and more users, different users will put forward more demands when using edge node services. In addition to providing rich computing power, how to make edge computing nodes play a more valuable role? Cloud network integration is a very important technical point of edge cloud. The integration of cloud network is mainly manifested in the following aspects: first, the requirements for location. The edge cloud service itself is insensitive to location. Although the customer scene has strong demand for location, it must have location attribute. Second, the cooperation between nodes is completely based on the network. At the same time, the edge cloud carries the MEC nodes of the carrier and the connection relationship between the central Region and the local Region. This connection, data flow and application path routing are realized through the integration of cloud and network technology. Every single node in the network can be considered untrusted or offline in a short time, which requires the integrated scheduling capability of the cloud network to ensure the high availability of edge cloud node services.
Ali Cloud has 2800+ edge cloud nodes all over the world to ensure that customer business can reach global users. Through the comprehensive cloud of CDN, CDN nodes are upgraded to edge cloud nodes so that customers can enjoy more edge computing services when using the original CDN resources. At the IaaS level there is the standard Edge Cloud node service ENS, which is a small distributed cloud node service. ENS base of edge cloud node provides unified resource construction, unified resource inventory, and unified external resource delivery capabilities, enabling various application scenarios to use ENS resources as unified operating systems. At present, ali Cloud edge cloud node service has built many application scene products, including CDN product system, cloud communication scene, view computing scene, cloud game and innovation scene jointly built with partners.
Ali Cloud Feitian strong technical ecosystem, cloud native direction for many years of accumulation and CDN ten years of a sword, more importantly, Ali cloud edge computing early industry layout and exploration — CDN on ENS to business platform, platform ecological strategic measures, created this Feitian edge technology map.
Ali Cloud edge cloud building capabilities and industry solutions
Ability to build | edge application hosting
We take the edge container platform built by the edge cloud IaaS base as the base. Based on the concept of complete cloud native, we build the Application publishing and control APaaS (Application PaaS) system that meets the cloud native standard and is no different from the central cloud experience. Based on the application management and control mode of cloud native OAM, it provides separation of concerns and lightweight application management platform services, and completes the cloud of edge infrastructure as a whole. Provides users with a real sense of EdgeNative edge application hosting services.
At present, many internal core businesses have been verified on a large scale based on the edge cloud native technology system constructed by us. Here, I would like to share with you a few representative scenes.
Industry solutions | CDN on ENS
The next generation cloud CDN constructed based on edge cloud ENS base. Based on the CDN cloud approach, we complement the shortcomings of isolation reprotection and resource reuse for traditional CDN scenarios, and also lay a solid foundation for business innovation. In addition, we broaden the business boundary of CDN based on the edge cloud native system, from the original Content distribution to the current definition of Computing Computing distribution network. It provides users with the core service capability of programmable CDN, which greatly optimizes the cloud cost for customers and improves the efficiency of business innovation.
Industry solutions products | flow calculation
Another typical scenario is the new product of traffic calculation. Based on the existing full-domain unified access scheduling capability of CDN, we built a powerful computing power network based on the edge cloud native base. It can deliver super-scale computing power to the whole network at the second level, and computing power can be used on demand. It has the consistent experience of the whole network, and supports payment by volume, which greatly improves the overall resource utilization efficiency and reduces the computing power cost of customers. Based on the wide coverage of Ali edge cloud, super-large edge infrastructure, and the unified resource scheduling and access scheduling capabilities of the whole domain, it truly achieves the movement of computing with the network to cope with various scenarios of edge computing power.
Industry solutions | digital twin and intelligent terminal on the cloud
For a thin client with playback capability but no computing power, the functions of rendering and strong interaction originally needed to be completed on the client are moved to the cloud. After rendering is completed on the edge cloud, the thin terminal can be used for single video playback and interactive data delivery.
Industry solutions | cloud game
In the field of cloud games, edge Cloud ENS solves the “last mile” of cloud games landing. Game application uploading completes the deployment of global edge cloud nodes and processes data requests nearby according to user distribution. Game distribution reaches millisecond level command effect, realizing fast distribution to the edge. Cloud games compute, analyze and issue instructions at the edge, truly achieving rapid business response.
future
The construction of edge cloud native technology system is from bottom to top. From the perspective of underlying infrastructure, IaaS, IPaaS, APaaS, edge service capability layer (EdgeMesh, edge application hosting, global scheduling, etc.) and edge capability open layer, it can be predicted that 5G edge computing is bound to be a symbiotic and win-win relationship, and the two cannot be separated. Especially in the case of more and more marginal scenes, and deeper and deeper into the industry and industry, all kinds of new needs and scenes are constantly emerging. From the large to the meta-universe, to the concrete vehicle-road cooperation and other vertical scenarios, higher requirements for edge calculation are put forward. Security, stability, regional energy tuning, cloud network integration, network following the cloud and other fields under the edge cloud native technology system are the direction of key breakthroughs. It will be our core work to effectively solve the problems of edge cloud and network based on the concept of cloud native.
The original link
This article is the original content of Aliyun and shall not be reproduced without permission.