Alibaba Cloud's "terminal cloudification" practice, building edge architecture based on ENS

The endless update iteration of the terminal is a surge in the software's demand for computing resources.

Author｜Editor by Wang Guangfang｜IMMENSE

Terminal cloudification: breaking the shackles of hardware

In recent years, "terminal cloudification" technology has begun to be implemented on a large scale. The core idea is "computing offloading", that is, the computing tasks of intelligent terminals are offloaded to edge cloud virtual terminals for processing, so as to solve the shortage of terminal equipment in terms of computing power and storage. .

For example, the recent popularity of cloud games has made it possible to play large-scale high-definition games on low-end mobile phones, PCs and even smart TVs.

In order to better understand the advantages and characteristics of terminal cloudification, the following will take Alibaba Cloud's view computing product VEC (Visual Edge Computing) as an example to illustrate.

View computing VEC focuses on cloud scenarios for video and image devices (consumer electronics, cameras, in-vehicle terminals, etc.), and provides services such as device management, view processing, and view storage through standard protocol access of device terminals.

The "terminal cloudification" technology breaks the shackles of hardware, and its advantages can be summarized as the following two points:

One is the decoupling of function evolution and terminal equipment. The evolution of functions and performance no longer depends on the configuration and upgrade of terminal equipment. Faced with various problems such as long cycle, poor consistency, and compatibility, only cloud services need to be upgraded uniformly.

For example, the ordinary camera of the VEC service needs to add the AI recognition function. There is no need to spend a lot of money to replace the high-priced smart camera. It only needs to connect the video stream connected to the cloud to the AI recognition service, and simply configure the parameters.

The second is efficient scale management.

When the scale of the terminal equipment reaches a certain scale, the management of the terminal itself and the management of the production data on the terminal will be a big challenge, especially the view equipment of the VEC service, which will generate a large amount of video or picture data, which is based on the platform in the cloud. The ability to perform device management, stream processing, and stream storage is beyond the reach of the device side.

View computing VEC, as a typical scenario of terminal cloudification, focuses on building the standardized protocol for one-click cloud access of view devices, streaming media processing such as screenshot/recording/transcoding/mixing, view AI, view storage and other core technical capabilities. Competitiveness comes from the architectural advantages built on edge cloud.

Why does "terminal cloudification" rely on edge cloud?

Edge cloud nodes are ubiquitous, closer to users, and have low data processing latency, which can bring a better experience.

What can edge cloud bring to "terminal cloudification"?

Still taking the view calculation VEC as an illustration, the conventional VEC system design can be a complete central architecture, which is relatively simple to build, as shown in the following figure.

It can be seen that the VEC system applications are uniformly deployed in one Region, providing access services of GB protocol and self-developed protocol for all view devices on the entire network, and supporting access devices to push streams with RTP or RTMP protocol for slice recording and screenshots , transcoding, AI analysis, storage, real-time playback, video review and other services.

However, VEC products serve a large number of view devices, and the view stream data needs to be continuously uploaded to the platform for processing. If all data are processed in a unified manner at the center, the cost is very high.

In such typical terminal cloud-based high-traffic scenarios, if traffic can be converged to the edge for processing, traffic costs will be significantly reduced, and product competitiveness will be improved.

At the same time, if devices on the entire network push streams to the center in a unified manner, it cannot ensure a low-latency experience of streaming from all devices, and it is very necessary to access streams from the edge.

At this point, two major advantages of edge architecture emerge.

One is the nearest access. The traffic gateway can cover a wide range based on edge nodes and support the nearest low-latency access of devices;

The second is lower cost. The device's view stream data can be processed and stored at the edge, optimizing traffic and storage costs.

How is a complete edge architecture system built?

Next, how to design the edge architecture of the business system?

The main difference between the center architecture and the edge architecture is the evolution from the "center-terminal" two-tier architecture to the "center-edge-terminal" three-tier architecture. As the middle layer, the edge cloud will focus on the upward movement of terminal capabilities and the sinking of central capabilities.

Upgrading terminal capabilities can reduce terminal hardware requirements and software complexity. For scenarios with a large number of terminals, there will be significant benefits in terminal cost optimization and easy operation and maintenance. This is also the value logic provided by VEC in cloud scenarios on view devices.

The ability to sink in the center mainly evaluates the modules in the business system that have cost differences between the edge and the center. For example, the view stream of VEC can be processed and stored at the edge after the edge is connected, and subsequent view data reading can also be directly accessed from the edge. In this way, the traffic from the edge to the center is greatly converged, and the bandwidth cost is optimized.

The evaluation of the central and edge modules of the VEC system can be distilled into the following categories:

• Global unified management and control modules need to be deployed in the center, such as device management, authority management, etc.;
• The difference between central and edge deployment is not obvious, and you can choose flexibly according to the scenario, or give priority to central deployment, and operation and maintenance are relatively simple, such as signaling gateways;
• For edge sinking scenarios, some modules need to be added to the center, such as node management, scheduling for edge multi-nodes, etc.;
• Service-based modules (view stream processing and storage related modules) try to sink to the edge, and perform regionally distributed services through scheduling, such as streaming gateways, stream processing, and view storage.

The final VEC edge architecture is shown in the figure below.

ENS is Alibaba Cloud's edge cloud product - Edge Node Service, which provides virtual machines, containers and other forms of instances, block storage, object storage and other storage services, as well as LB, NAT and other networks. components, to complete the global coverage of key cities across the country.

VEC builds an edge architecture based on ENS, which significantly reduces the research and development cost of technical solutions. At the same time, the product-level guarantee of the stability of the edge infrastructure layer can focus more on the technical iteration of the business system itself.

For the edge architecture, the system needs to build some key capabilities to better utilize the advantages of the edge architecture, including:

Resource and business scheduling capabilities

The resources are upgraded from the central unified model to the edge distributed model, and supporting resource and service scheduling become the core capabilities of the system, including access and traffic nearby scheduling, and availability scheduling when node failures or high water levels are present;

Cloud-side collaborative management and control capabilities

Central control and edge control together form a closed loop. For example, a user arranges a processing process through the central OpenAPI, first slices and records, stores it as a fixed-length file, and then reads the file for AI recognition, which involves the central parsing template -> multi-module parameters Sending edge -> edge module processing according to process and parameters -> identifying structured information back to center -> center information processing and other complex links of cloud-edge interaction management and control;

Stable and secure communication capabilities on the cloud edge

Internet-based remote communication RPC between the center and the edge needs to focus on communication security and communication reliability, and the cloud-edge collaborative management and control relies on the cloud-edge stable and secure communication capability;

Edge autonomous management and control capabilities

The edge needs to have certain autonomous management and control capabilities to meet the requirements of high-performance, high-availability, cloud-edge consistency and other scenarios, such as reducing the dependence on the center in edge processing logic, and when cloud-edge communication is unavailable for a short time. State synchronization etc. after recovery.

What else is needed to build a complete edge architecture system? supporting system capabilities.

mainly include:

Edge Nodes Efficiently Deploy Edge

Node deployment includes application and initialization of node resources, as well as deployment and release of edge module applications. The edge architecture requires a large number of edge nodes for services. The deployment and management of nodes must be standard and efficient. Application containerized deployment is a good choice.

Configuration management and configuration dynamic distribution

Core capabilities such as business scheduling and cloud-side collaborative management and control all require dynamic management based on policies and configurations, enabling flexible scenario adaptation and more efficient use of network-wide resources.

Remote operation and maintenance system

The remote operation and maintenance capabilities include the collection and reporting of edge status and logs, monitoring and alarming, remote diagnosis and disposal, etc. The instability of the Internet increases the challenges of remote operation and maintenance. The links for operation and maintenance actions are long and there are many failure factors. Build full-link capabilities in dimensional scenarios.

The big picture of VEC's support system is as follows, in which key capabilities such as application deployment and release, service discovery and governance, and cluster expansion and contraction are built on the ENS container platform.

The development of edge cloud has attracted the attention of more and more technical people. While paying attention to and participating in the evolution of edge cloud capabilities, it is believed that what more technical people have the opportunity to do is to promote the evolution of their business architecture to edge architecture. .

The view computing architecture can be extended for more terminal cloudification scenarios and evolve into a more general terminal cloudification platform.

In terms of platform technology, the currently self-developed ODCAP (Open Device Cloud Access Protocol) open device cloud protocol can be used for a wider range of terminal cloud access and management, and a visual view stream processing and orchestration engine is provided to facilitate terminal data. complex processing flow definition.

In terms of edge architecture evolution, view data is stored in a large number of edge nodes, and data operations need to locate nodes before reading and writing, and the complexity is high.

Technically, it should be considered to abstract the storage of multiple physical nodes into a logical storage, and encapsulate a set of standard logical object storage interfaces, which can be completely consistent with the central object storage interface in terms of user experience. This capability has also been incubated on Alibaba Cloud Edge Cloud ENS, and it has begun to provide external product services.

With the continuous enrichment of edge scenarios, edge cloud can achieve better development. The mission of edge cloud is to support better, faster, and more stable construction and development of various edge scenarios and architectures.

In short, really good technology can break the shackles and bridge the gap.

"Video Cloud Technology", your most noteworthy public account of audio and video technology, pushes practical technical articles from the frontline of Alibaba Cloud every week, where you can communicate with first-class engineers in the audio and video field. Reply to [Technology] in the background of the official account, you can join the Alibaba Cloud video cloud product technology exchange group, discuss audio and video technology with industry leaders, and obtain more latest industry information.