Application serverless

On November 3, at the 2022 Hangzhou Yunqi Conference, Alibaba Cloud President Zhang Jianfeng said that a new computing system with the cloud as the core is taking shape, and the software research and development paradigm is undergoing new changes. Serverless is one of the most important trends. Alibaba Cloud will firmly promote the comprehensive serverless of its core products to help customers better achieve agile innovation.

On November 5, we activated a new paradigm for building applications: the Cloud Native Summit again focused on Serverless, further interpreted the significance of Alibaba Cloud's comprehensive serverless core products, and launched a major serverless runtime upgrade to make building applications on the cloud easier.

AliCloud Smart Observable&Keynote Speech by Si Tufang, Head of Serverless

Serverless leads the next generation application architecture

To talk about the application of Serverless architecture, we should first understand Serverless from the evolution process of cloud product hosting.

"Infrastructure hosting" is the most basic form. Products deliver cloud resources such as computing, storage, and network. Users need not only to deploy and maintain application software and business logic on the cloud, but also to solve various problems that may be encountered in software operation.

"Application hosting" is an upward extension of "infrastructure hosting". Customers still need to purchase cloud resources in the mode of "several cores and several gigabytes of servers", but the products provide part of the operation and maintenance of common applications (such as MySQL).

Serverless full hosting is a further evolution. Customers no longer need to care about the server. The server is fully managed by the cloud product and has two important features: first, it is paid according to the actual usage, which is closer to the "grid" mode: for example, it is called according to the number of requests, or paid according to the actual data storage; Second, adaptive flexibility and operation and maintenance free: according to the usage, cloud products automatically scale the underlying resources. Customers do not need to pre order resources in advance, and they can recycle them when they are used up.

The appearance of Serverless full hosting is deeply affecting and changing the application technology architecture. From the enterprise application architecture to the Internet distributed architecture, the concepts of service-oriented, scalable, loose coupling, etc. have been deeply rooted in the hearts of people, but the implementation complexity of distributed technology continues to rise. The self adaptation and maintenance free features of Serverless can greatly reduce the complexity, and its high elasticity gives full play to the advantages of the cloud. Relying on Serverless fully hosted products, business capabilities and cloud service capabilities can be abstracted into flexible and universal module forms. Users can select modules from them as needed, adjust them as needed and arrange them together to assemble their own applications, thus greatly improving the R&D efficiency.

From resources to services, Alibaba Cloud's core products are all serverless

Serverless does not use a server. Instead, it fully trusts the server to the cloud manufacturer. It automatically scales according to the size of business traffic. Out of the box, it eliminates maintenance costs and is billed based on usage. Users do not need to care about and manage the underlying IT resources. They just focus on business code and process business according to actual requests.

To enable users to make good use of Serverless, it is far from enough to simply implement serverless at the application runtime level. If the downstream databases and other systems that the application depends on do not have good flexibility, they will become the "weak board" of the system as a whole. Only when all systems on the link are highly resilient and reliable can Serverless maximize its value.

At present, more than 20 core products on Alibaba Cloud have provided the Serverless form, which helps customers better control the underlying computing power and save costs in terms of elastic speed and billing model; At the same time, around the close cooperation between serverless architecture products, we will work together to solve the problems of product integration, linkage scaling, and so on, so as to provide a more smooth full link serverless experience for the business.

With the growing popularity of Serverless, Alibaba Cloud has launched the Serverless tool chain and the Serverless application center to help developers improve the online end-to-end productivity of serverless applications. At the same time, Alibaba Cloud has also accumulated a large number of serverless solution cases for reference by all walks of life.

In 2020, Alibaba Cloud's open-source Serverless Devs will become the industry's first cloud native lifecycle management platform that supports mainstream Serverless services/frameworks. This year, Serverless Devs was deeply involved in the formulation of the standard of "Capability Requirements for Tool Chains Based on Server less Architecture" of the ICT Institute, and officially entered CNCF Sandbox in September, becoming the first Serverless tool project selected in the industry.

Alibaba Cloud's Serverless computing product

Alibaba Cloud is the first cloud manufacturer in China to provide Serverless computing services, and its product matrix is also very rich. According to the different abstraction levels, the elastic container instance ECI, Serverless K8s ASK, Serverless application engine SAE, and function calculation FC correspond to the container, cluster, application, and function level abstractions respectively.

• Elastic container instance ECI. This product provides container instances. Developers do not need to manage underlying servers, but only provide packaged images to run containers. Strictly speaking, it is not serverless. The upper layer needs to solve the problem of how many containers are needed for how much load, when to expand and when to shrink, and how to schedule traffic.

• Serverless container service ASK, which belongs to the container orchestration layer and can perform ECI node orchestration and capacity planning. ASK can help set up clusters and reduce the threshold for Kubernetes to use. It only needs to manage network segments, load balancing, applications, etc., rather than the underlying node infrastructure.

• Serverless application engine SAE, the first application oriented Serverless PaaS platform in the industry, shields the complexity of the underlying IaaS and Kubernetes, provides a zero code transformation and migration, more cost-effective and more efficient application hosting solution, and helps users achieve serverless single Web applications, microservice applications, and scheduled tasks.

• Function computing FC, a FaaS product, is an event driven fully hosted computing service. Users only need to write and upload code, and function computing will automatically prepare computing resources, run code in an elastic and reliable way, and provide complete observability, greatly simplifying the development, operation and maintenance process.

Serverless runtime upgrade: easier to build cloud applications

Serverless application center: make serverless easier to develop

Serverless has become the core technology of cloud computing since its development. The mainstream scenarios are all solved through Serverless. Alibaba Cloud has provided a complete tool chain, allowing enterprises to build applications on the cloud more easily through the Serverless architecture, and fully enjoy the dividends of serverless.

With the popularization and use of the Serverless architecture, problems such as the imperfection of the Serverless tool chain system, the complexity of the update/deployment process, the fragmentation of resources and the difficulty of governance have also emerged. Alibaba Cloud has launched the Serverless application center with a massive number of scenario based templates, making the full lifecycle management of serverless applications easier. By using the Serverless application center, users can quickly deploy and manage applications without additional cloning, building, packaging and publishing operations before deploying applications, helping users quickly link upstream and downstream services on the cloud and easily precipitate best practices.

The Serverless application center further covers the whole life cycle of Serverless applications from creation, development and operation and maintenance, including white screen experience, cloud development, application management in multiple environments, automatic resource preparation of application dimensions, standardized DevOps processes and other enterprise level features.

The subsequent application center will continue to deposit the typical Serverless application case templates from all walks of life, so that users can understand and master them more easily. At present, the application center has added more than 100 application templates for nine scenarios, including event driven, Web API, and audio and video processing.

Function calculation FC: flexible and high-energy, broaden three scenarios

Since its release, FC has helped tens of thousands of domestic and foreign enterprises to implement modern applications in a wide range of scenarios, such as the Web, mobile backend, audio and video, AI reasoning, and batch task processing. This year, FC has penetrated three scenarios, including audio and video processing, real-time data processing, and GPU, to help developers focus on business, reduce costs, and improve efficiency.

In terms of audio and video processing, FC has added a panorama recording template that can be used directly out of the box. Through the WYSIWYG mode of panorama recording, it can easily restore the live interaction effect, which can not only provide a SaaS audio and video fast access experience, but also have code level customization flexibility. With strong flexibility, FC can instantly create multiple instances for video multi-channel parallel transcoding, which greatly shortens the film production time. At the same time, FC has high resource utilization and low computing power consumption, which reduces the cost of traditional schemes by more than 70%.

Let messages flow, and FC's real-time data processing capability is further enhanced. FC has official integration with all the messaging products Kafka, RocketMQ, and EventBridge, and hundreds of trigger sources are built in to enable one click data cleaning, enrichment, and dump of messages on the console, so that messages can flow and play a greater role. FC's adaptive elasticity can effectively cope with the peaks and troughs of massive messages, reaching the event throughput of 100 million per minute.

The hardware computing power represented by GPU has gradually replaced the traditional CPU and become the computing power provider of AI reasoning and multimedia processing. Through our test, the performance of these scenarios using GPU can be improved tens or even hundreds of times as much as that of traditional CPUs. However, the price of GPU is not cheap, and the utilization rate is generally lower than 30%. Many enterprises either "cannot afford" or "cannot use" GPU. In response to this pain point, FC introduced Serverless GPU, which supports multi specification GPU computing power division with a minimum of 1/16 card. At the same time, it provides quasi real-time cold start, second level flexibility and second level billing, making GPU computing power more affordable and benefiting SMEs.

Serverless application engine: new loads, new scenarios, new tools

As the first application oriented Serverless PaaS product in the industry, the Serverless application engine SAE has become the first choice for enterprises to go to the cloud and use the cloud with the characteristics of low threshold microservice+container transformation, high flexibility and no operation and maintenance. It has always been the product philosophy of SAE to enable popular open-source architecture workloads to be directly serverless. At present, there are tens of thousands of enterprises' Spring Cloud and Dubbo microservice applications on SAE. Serverless can be completed without a single line of code.

Continuing the same concept, SAE has added job type workloads this year. Open source K8s Cronjob, XXL JOB, and Apache ElasticJob can all be hosted seamlessly. SAE can quickly release computing resources after task completion, with lower cost, and has additional capabilities such as failure retry, parallelism, sharding, and built-in observability.

In addition, SAE has also expanded its support for multilingual microservices. Whether PHP, Python or Go, services can be registered and discovered based on SAE. The most common multilingual gRPC protocol is also supported. Relying on the industry-leading eBPF technology, SAE can provide non-invasive multilingual application observability, and provide common indicator insight and call topology view for container instances, applications, services, etc.



The tool chain is related to the enterprise's R&D operation and maintenance process, as well as the continuity of developers' habits. This year, SAE has further enriched its support for enterprise common tool chains: integrating command line tools and script systems through Serverless Devs, integrating "Configuration is Code" (IaC) systems through Terraform, and integrating continuous integration (CI) systems through Jenkins plug-ins.

Overall price reduction: make Serverless a cloud based hydropower coal that benefits the public

Function computing ushered in a comprehensive price reduction, with the unit price of vCPU dropping by 11%, and the maximum drop of other independent billing items was 37.5%.

The FC billing granularity of the function calculation is fine. The billing is based on the memory and execution time of the execution environment. The billing granularity can reach the millisecond level. The user can only pay for the resource consumption generated by the request. The minimum billing rule can reach the billing duration of 1 millisecond granularity, 0.05 core vCPU, 1/16 GPU card. The binding of each billing item, such as function specification vCPU, memory, disk, etc., is completely unbound, allowing users to freely choose the specification according to their own application runtime overhead, and further reducing the proportion of idle resources. The reservation mode is the ultimate way to eliminate cold start for FC as a function of calculation. The idle charging is only 1/10 of the price, which allows the business to avoid the worries of high costs.

According to the team's estimation, if the daily average utilization rate of a cluster's resources is below 30% or there is obvious idle waste, it is suitable to use the function to calculate FC. After FC is calculated by function, the resource utilization rate can be increased to 60% or even more than 90%, and the comprehensive cost can be reduced by 15% to 70%.

Take some enterprise use scenarios as examples:

Video live broadcast is the most important business of love social APP, with obvious peak valley characteristics. In order to ensure the stability of the business, needs to prepare a large number of machines to process tasks. During the low peak period of traffic, machine resources will be largely idle and wasted, while the peak brought by some holidays will exceed the maximum capacity of the cluster. The task queue has to downgrade some businesses. After using the function to calculate FC, the resource problem in peak and valley periods is completely solved, and the resource cost expenditure is reduced by 20%.

Yuchuan Technology, an Internet marketing and promotion service provider, is mainly based on the Alipay applet. It has the characteristics of large fluctuations in access volume and difficult to predict traffic bursts. In particular, the sudden increase in access during the event is also a great test of the stability and flexibility of the applet back-end services. In order to deal with unpredictable sudden traffic, Yuchuan Technology has implemented the whole system serverless. It can support small programs with daily activity of more than 500000 people for only 200 yuan a day, and can withstand tens of thousands of QPS.

The comprehensive price reduction will make Serverless the universal hydropower coal, which will be the first choice for innovative enterprises to go to the cloud.

Serverless power behind thousands of industries

Perhaps for many people, Serverless is still a new concept, but when you send out a microblog, click on a song, enter a live broadcast room, and pay the bill in the supermarket, you may have the silent support of Serverless.

The technical scenarios covered by Serverless are constantly expanding. At present, Serverless has been verified and widely used in many scenarios such as microservices, online applications, event driven, and task processing. If your business traffic is unpredictable, or there are tidal fluctuations, or there are obvious idle resources, and the resource utilization rate is less than 30%, Serverless will be a good tool for you to reduce costs and increase efficiency.