On June 9, 2021, the asia-pacific conference on content distribution and CDN summit to be held in Beijing, ali cloud intelligent edges Li Zhongren cloud senior technical experts invited to accelerate the BBS, share the ali cloud based on the edge of cloud nodes to build a new generation of CDN product ability, the technical architecture and innovation practice, help enterprises improve efficiency and optimize operation, We will promote the innovative development of more emerging industry scenarios. The picture
Li Zhongren introduced that Ali Cloud CDN originated from Taobao CDN and grew gradually with Double 11 since 2009. It provides accelerated support for e-commerce websites with huge traffic and has established hundreds of high-quality nodes and bandwidth resources around the world. At first, it has been applied and practiced in the field of e-commerce. In 2011, CDN, as an Internet traffic distribution infrastructure, gradually expanded from serving Taobao e-commerce to comprehensively serving Ali Group ecology, and extended in application scenarios. In 2014, Ali Cloud CDN officially announced its commercialization, opening its service capabilities to global enterprises and entrepreneurs. In 2015, Ali Cloud CDN and e-commerce joined hands to stride forward the whole site HTTPS era, and continuously carried out the iterative upgrade of its own scheduling system to cope with the impact of exponential growth of traffic. With the annual Double 11 and the popularity of mobile Internet and video, CDN traffic shows exponential growth. In 2017, Ali Cloud CDN launched the globalization strategy, and released SDCN, DCDN and other products successively, comprehensively upgrading the product matrix. In the Market Guide for CDN Services released by Gartner at the beginning of 2018, Ali Cloud was designated as a Global (Global level) service provider by virtue of its Global service capabilities and product technology advantages. In recent years, Ali Cloud CDN has continuously carried out in-depth research on product technology centering on “intelligence” and “computable”, and continued to polish on edge Serverless, multi-dimensional Workload load balancing, scientific decision-making, and various edge application scenario services. At the same time, Ali Cloud CDN has won the trust of customers by virtue of more efficient products and services and high-quality service experience. Its volume in the CDN field keeps expanding and developing rapidly, and the peak traffic scale of CDN keeps setting new highs. Up to now, Ali Cloud CDN has 2800+ edge nodes and 150+T bandwidth reserve in the world, serving hundreds of thousands of enterprise customers around the world and providing accelerated services for over one million domain names. Ali Cloud CDN creates an edge ecological network connecting global end users, processing hundreds of millions of QPS user connection requests per second during the evening peak, and delivering billions of configuration management instructions to all nodes of the whole network every day.
CDN industry development and evolution picture
The development of CDN has gone through three major stages: the first stage is dominated by the consumption of graphic and text content, people mainly use PC to surf the Internet and visit portal websites and other content containing a large number of graphic and text;
The second stage is dominated by graphic and video content consumption. The popularity of live broadcast, short video and long video leads CDN into a period of rapid growth, and the bandwidth increases exponentially every year. But the second phase of explosive growth has passed, bandwidth growth has slowed, and the second phase is coming to an end.
The third stage is now in the gestation period. With the development of 5G and IoT, Internet of Things devices are widely popularized, more and more devices will be connected to the network, and there will be iconic applications to drive the great development of the whole industry. The distribution of CDN will also expand from the distribution of single bandwidth to the distribution of bandwidth, computing power, storage and other multidimensional distribution.
Image of new CDN technology architecture based on edge cloud node
When talking about the concept of new CDN, Li Zhongren said: CDN is transformed from traditional content distribution into multi-dimensional distribution such as bandwidth, computing power and storage. The technical base of CDN has evolved to the edge cloud operating system. Traditional CDN is only an application on the edge cloud, which mainly consumes bandwidth. In addition, many applications that consume computing power and storage will run on the edge cloud operating system. Ali Cloud provides the underlying operating system. Users can construct their own operators and host them on the edge cloud nodes of Ali Cloud according to their own business conditions. Aliyun provides users with automatic cloud services such as nearby access, nearby computing, location-insensitive, on-demand deployment and elastic scaling. This new service form is complementary to the advantages of traditional CDN. It not only reuses the uplink bandwidth and idle computing power of CDN, but also provides users with less delay and better cost services.
Picture of XDN overall architecture
Li Zhongren pointed out that the overall FRAMEWORK of XDN evolved from ali Cloud’s technical accumulation and precipitation in the CDN field, which is divided into resource delivery layer, platform layer, component layer and product layer. The resource delivery layer builds wide-area coverage edge cloud computing infrastructure based on operator resources. The hardware and network composition of nodes are heterogeneous, and the supply chain needs to ensure efficient access of various heterogeneous resources to the production system. The platform layer connects the upper business with the bottom resources, and integrates the supply chain capabilities to build the XDN PaaS platform, which includes mid-stage support, intelligent scheduling and network protocol. The mid-stage support system includes monitoring, logging, service management and control, and resource management, which involves massive data processing, real-time delivery of piping instructions, service and resource management, billing, and operation status monitoring, etc. Intelligent scheduling includes service orchestration and resource scheduling. It mainly solves the problem of making automatic decisions about which services run on which resources and which resources should run on services. Together with global load balancing, the elastic capability of XDN system is determined. Component layer, with high-performance access gateway, high-performance cache server, high-performance 4-layer cluster load balancing and other traditional CDN basic component functions; And programmable CDN, function calculation, edge transcoding, streaming media processing and other ecological epitaxial functions. Nodes covered by a wide area undertake massive services, requiring a system of security protection policies to protect against attacks, protect content data, and ensure transmission reliability. The top layer is the product output layer, which includes CDN acceleration, content download, live broadcast, on-demand, SCDN, DCDN, EdgeImage, EdgeRoutine, view computing and other new products to the edge cloud.
The difference between the new CDN and the traditional CDN
The main difference between the new CDN and the traditional CDN lies in whether users can customize operators and upload them to edge cloud nodes to enjoy on-demand deployment, location-insensitive and flexible scheduling services. Compared with bandwidth-based services, computing power consumption has its own characteristics and difficulties: Computing power consumption needs to be accurately evaluated: bandwidth consumption can be accurately evaluated through access logs, but computing power needs to be accurately evaluated because many services share CPU on host computers.
Computing power has higher requirements on the accuracy of scheduling: Bandwidth If inaccurate scheduling leads to high node water levels, the impact on services is small, and network packet loss can be almost no service through retransmission mechanisms. However, if the CPU water level is too high, the whole machine is almost unserviceable, which greatly affects services.
Ali Cloud has realized the upgrade of global intelligent scheduling system, which has evolved from scheduling based on bandwidth dimension to multi-dimensional resource scheduling model driven by bandwidth computing power. The input data of this model are bandwidth consumption and computation power consumption data. For computing power consumption data, the current node services and the current CPU consumption to do business portrait to solve. Finally, Li Zhongren introduced innovative scenarios based on edge cloud nodes in detail: ER, EI, view computing, GRTN.
Edge Innovation Scene -ER image
ER, short for EdgeRoutine, is a JavaScript code runtime environment that runs on CDN edge nodes. Developers can upload JS code to ER, so that JS code can run on CDN edge nodes around the world. It is equivalent to having thousands of microservers around the world to serve users around the world (clients). The CDN for simple cache distribution is equipped with computing capability, and the small traffic service becomes a scenario with high added value. Traditional CDN request chain: after arriving at Tengine, go to Swift to find the cache, if the cache name is directly taken from the cache to the client, if there is no hit, go back to the source to request the content. In ER mode, there are four modes: First, after the request reaches the gateway, it is directly processed by ER. After the request is processed, it is directly returned to the customer through the gateway without being cached.
The second type of ER makes sub-requests to other public network services, such as OSS, and then returns the results to the customer through the gateway.
After the third request reaches ER, it can read or write the calculated result from the cache and KV storage. The result of one calculation can be reused by a large number of other requests to reduce unnecessary repeated calculation and delay.
The fourth is the most complex. ER and CDN services are coupled. After the request is sent to the gateway, ER preprocessing is first entered, and then the subsequent CDN cache and source logic are followed, and the response is returned to the customer through ER and gateway.
When using ER, you do not need to worry about resource shortage. Intelligent dispatching automatically scales based on service volume to meet service requirements. By means of computer isolation technology, tenants are isolated in a process, resources are utilized to maximize utilization, idle computing power is utilized at the edge, and optimal pay-per-use services are provided to users. Users only need to care about their core business code, write the code and upload it to Ali Cloud to complete the deployment, do not care about which machine the code is running on, Ali Cloud provides you with the guarantee of the underlying operating environment. At the same time, Ali Cloud covers more than 2800 nodes around the world, ensuring that no matter where customers are, they can always find the edge cloud nodes close to them, thus greatly shortening the network link, reducing the risk of network congestion, and providing low-delay services.
Edge Innovation Scene -EI Image
EI is Edge Image, which is the Edge Image processing function launched by Ali Cloud. Through this function, CDN can directly process and distribute the Image row at the source node, which can reduce the source station pressure, reduce the source link and save the source traffic. With the image processing function, customers can scale, cut, add watermarks and other operations on the original image on the CDN to meet the image processing requirements of various business scenarios (such as e-commerce, media, games, finance, etc.). In the overall architecture diagram of EI, the image conversion service is first split into containers, so that business isolation can be achieved, and other services will not be affected by their own process problems. In addition, independent control of resources can be achieved. As a picture conversion service, EI will be deployed on THE CDN L2 node as required and take effect in the back source domain rather than in the access domain. In this way, the number of picture conversion can be reduced and computing power can be saved. One conversion can take effect on multiple edge nodes. The image conversion request is intercepted and forwarded to EI container by bypass for image conversion. At the same time, EI supports hot restart, the upgrade process will not have any impact on the ongoing conversion. All target images converted from the same image are copies of the original image. Refresh the original image once, you can brush out all the corresponding target image. CDN has a limit on the number of times of refreshing submitted by users, which can greatly save the number of refreshing times for users with multiple target graphs. CDN will cache the original map when the original map is transferred to the target map. When turning to another target graph, the CDN can turn to another target graph without returning to the source. In order to reduce access to the source, customers typically preheat resources in advance. For a scene where one image corresponds to multiple target graphs, the original image only needs to be preheated once, and the original image will be taken from THE CDN itself when multiple target graphs are transferred to reduce the back source. CDN image conversion can save resources for the source station. The source station only needs to save the original image, reducing the cost of storing multiple target images and reducing the cost of transcoding calculation. In the case that the original image is cached, the overall link of image conversion is reduced by one link, that is, the section of CDN returning to the source station, which reduces the overall request time. In general, the user needs to modify the request by adding parameters to specify the specific transformation operations to be performed. Combined with its own capabilities, CDN can provide some functions that users can convert pictures without modification: adaptive WEBP
According to the request header Accept, CDN determines whether the current request client supports webP format. For example, it supports automatic conversion to WebP format and then returns it to the client. Webp format due to its high compression ratio, for example, compared with JPG format, the same resolution, the size will be smaller compression, saving traffic.
A key thin body
Can be configured to a certain format of the image request unified quality scaling. In the naked eye can not see the change of the premise, reduce the quality of the picture, to save bandwidth for customers.
Adaptive quality of weak network environment
CDN can judge if the network speed is slow according to the network environment of the request, and then return the image after compressing the quality to improve the success rate and response time.
Edge Innovation Scene – View computing images
CDN mainly consumes the downstream bandwidth, while view services can reuse the upstream bandwidth and idle computing power of CDN, which just complement each other. Nowadays, we can see all kinds of cameras everywhere in our life, which means that we need to connect to widely distributed edge nodes for computing and storage. Most of the work will be done on the edge, with a few scenarios requiring the edge processing to be uploaded to the center. View to calculate the whole link is divided into three parts, respectively is the production side, calculation of end, consumption, ali cloud provides service mainly in the middle of the calculation of end, include: the view of processing (recording, transcoding, etc.), the view of storage (stored locally, and later returned to the center Region), view, access and control (access to support different terminal equipment). At present, view computing is mainly applied in intelligent transportation, intelligent education, public security and other fields.
Edge Innovation Scene -GRTN image
GRTN is a global real-time transmission network, which is committed to building a communication level streaming media transmission network with ultra-low delay and full distributed sinking through heterogeneous nodes of edge cloud. Compared with the traditional live broadcast architecture, the technical characteristics of GRTN architecture are as follows: hybrid networking: tree hierarchy structure + peer graph network capability sinking; protocol edge unloading + internal transport protocol normalized control and data separation; dynamic path planning + fully distributed SFU programmable and computable: The core values brought by the upgrade of streaming media Serverless computing service architecture are as follows: Excellent cost. GRTN is a multi-service integrated network that can support multiple scenarios such as live broadcasting, RTC and video cloud, with high service reuse rate. In addition, GRTN has shorter internal links and lower costs within nodes.
To improve quality, GRTN internal network supports the network structure constructed by dynamic route selection. The internal link delay can reach about 20ms, and the internal link adopts private protocol for efficient transmission. In addition, the push stream and distribution of the client are built based on WebRTC. QoS congestion control is designed specifically for streaming media characteristics, and it is continuously iterated and polished based on online data construction.
Easy to expand, GRTN supports WebRTC protocol, and can carry out full duplex communication on a single connection channel, so that it can publish and subscribe media streams freely, bringing more imagination space in service scalability.
Taobao live has fully used GRTN, and smooth guarantee taobao 618 live promotion, double eleven promote live and other scenes, and the traditional RTMP/ HTTPFLV way compared to the user experience, RTS end-to-end delay reduced by 80%, the rate of delay reduced by 30%, driving taobao GMV improvement. In addition to Taobao Live, GRTN serves customers in e-commerce, education, pan-entertainment and other industries.