Difficulties and Breakthroughs of Serverless Landing

It is just ten years since Serverless was proposed in 2012 to 2022 this year.

In the past ten years, cloud access has been a deterministic trend. At this stage, enterprises focus on how to achieve smooth cloud access. As more and more enterprises go to the cloud, and even many enterprise systems are built on the cloud on the first day, the core focus of enterprises is to make better use of cloud capabilities.

In the new stage, the role of cloud services has also changed from simply providing resources to a new platform for enterprises to build applications, helping enterprises minimize low-value repetitive work such as machine operation and maintenance, and focusing on business innovation. The core capability of cloud vendors is to provide serverless cloud services for enterprises, eliminate the complexity of cloud use, improve resource utilization in a simpler way, and help enterprises reduce costs.

In recent years, not only cloud vendors, but also a large number of open source commercial products have adopted the Serverless model, including Confluent Cloud, MongoDB Atlas, Snowflake, Databricks, etc.

As cloud manufacturers launch more and more serverless services in storage, computing, middleware, big data and other fields, the cloud product system is gradually serverless, which also drives the serverless architecture from the concept to the stage of large-scale production. In 2021, DataDog released the Serverless research report. The data shows that from cloud-based startups to large enterprises are paying attention to Serverless, and the Serverless ecosystem has surpassed FaaS, including dozens of services, which can help developers build faster and more dynamic applications.

The Singularity of Serverless has come. The so-called Singularity is the turning point from stable development to high-speed development, which also means that the landing of Serverless in the whole industry will be fully erupted.

However, at the same time, we also see that on the one hand, the attention of enterprises is increasing and the industry is beginning to land. On the other hand, enterprises and developers also have some confusion and concerns about Serverless.

Alibaba Cloud, together with InfoQ, collected the feedback of some enterprises and developers on the current stage of Serverless, and held the offline salon of the Cloud Native Serverless Developer Evaluation Bureau on December 16, inviting Alibaba researcher, Alibaba Cloud Intelligent Cloud Native Application Platform leader Ding Yu, Alibaba Cloud Serverless R&D team, Shanghai Huiyou Culture, Deji Network Technology Co., Ltd., Pumpkin Film, BMW and other enterprise technical leaders, We will discuss the challenges and experience in the process of All on Serverless together, with a view to providing reference and reference for more enterprises that implement Serverless. This article intercepts the wonderful speech on the salon.

See Alibaba Cloud native video number for the full version of video ↓

What is Serverless?

Before we talk about Serverless, let's first understand what Serverless is? Cloud and cloud origin

, like 17

In 2012, the concept of Serverless officially appeared on the stage. In 2016, the first Serverless ConF pushed the discussion of this technology to a peak. In 2017, Alibaba Cloud launched the first heavy serverless product service in China - Function Compute (FC), which allows developers to focus on business logic development instead of spending a lot of energy on the management of servers and other infrastructure, so as to quickly realize business value.

From 2018 to 2019, Berkeley published two consecutive papers defining Serverless, predicting that Serverless will become the most important computing paradigm in the cloud era, and Serverless has gained more and more attention.

Serverless is a very broad concept, not limited to computing. Generally, the Serverless service can be called if the following conditions are met simultaneously.

First, full custody service. It means that enterprises use abstract service interfaces instead of directly facing the underlying resources, so they have no burden of installing, configuring, maintaining or updating software and hardware. The full custody service usually also provides built-in fault-tolerant, safe and observable capabilities, and users usually do not need to rebuild these capabilities.

Second, adaptive elasticity. It means that the service can automatically scale flexibly according to the size of the load, which not only enables the system to support the rapid development of business, but also greatly improves the efficiency of resource use.

Third, pay according to actual usage. It means that you only need to pay according to the actual execution time, traffic or number of calls, which reduces the cost.

Therefore, the core value of Serverless service is to eliminate customer non-functional code development as much as possible, simplify the work of application infrastructure management, and achieve a leap in research and development efficiency.

Song Jie, CEO of Shanghai Huiyou Culture, experienced the first Serverless Conf. In his view, Serverless is a very subversive technology and solution, which is fundamentally different from all previous technological innovations, because the core issue involved is the change of business model. "It is similar to the leap from machine language to high-level language, which simplifies the problems that programmers need to care about when they are doing enterprise applications, and greatly reduces the complexity."

Serverless leads the next generation of application architecture

The Serverless service is in line with the trend of the evolution of the R&D model. In the past decade, major cloud manufacturers have been continuously making the product system Serverless.

As the first Alibaba Cloud to promote the implementation of Serverless in China, the evolution process of Serverless is also the development process of cloud computing. Ding Yu, a researcher at Alibaba and head of Alibaba Cloud's intelligent cloud native application platform, mentioned in the sharing that Alibaba Cloud is the first enterprise to do cloud computing in China. Since its establishment in 2009, the first stage has started with resources and IaaS. At this stage, many technologies have been open source and become popular on a large scale, including big data, AI, middleware and other micro-service and application-layer technologies.

From the perspective of cloud, cloud vendors and open source have a very good synergy at this time, so there are many fully managed cloud products on the cloud. From the perspective of enterprise customers, to ensure the reliable and efficient operation of applications on the cloud, the complexity of the cloud is very high. At this time, enterprises enter the stage of wanting to better use the cloud. For the evolution of the cloud, it also starts to evolve to a service-oriented way, that is, the cloud starts to gradually change from providing resources to providing services.

In this context, Alibaba Cloud has gradually moved towards the form of serverless services, including the birth of serverless products such as computing, storage, data, database, big data and even audio and video.

Alibaba Cloud Serverless evolution and practice

In 2017, Alibaba Cloud launched the FaaS product function computing FC, which is a fully managed computing service with event-driven as the core. Users only need to write and upload the code, and the function computing will automatically prepare the computing resources, run the code in a flexible and reliable manner, and provide complete observable capabilities, greatly simplifying the development, operation and maintenance process. In the second year, Alibaba Cloud launched the Serverless application engine SAE, which is the first application-oriented Serverless PaaS platform in the industry, shielding the complexity of the underlying IaaS and Kubernetes, providing an application hosting solution with zero code transformation, better cost and higher efficiency, and helping users achieve Serverless of individual Web applications, microservice applications and scheduled tasks.

In the same year, the leading industry launched the Serverless container service ASK, which is based on the elastic container instance ECI (Elastic Container Instance) to expand the capacity of 2000 pods in one minute, reducing the threshold of Kubernetes use, and allowing users to focus more on applications rather than managing the underlying infrastructure.

In 2020, Alibaba Cloud will open source Serverless Devs, becoming the industry's first cloud native lifecycle management platform that supports mainstream serverless services/frameworks. In September 2022, the project officially entered CNCF Sandbox and became the first selected Serverless tool project in the industry.

Serverless makes the granularity of cloud computing very fine. It is not a simple software problem for enterprises to choose whether to adopt Serverless products. Serverless needs a large enough computer cluster and product system capability to ensure its flexibility to meet the complex business needs of enterprises. Alibaba Cloud provides a serverless product portfolio at all levels, from container to application, and firmly promotes the comprehensive serverless product system.

At present, more than 20 core products of Alibaba Cloud have achieved Serverless. The number of FC calls of Alibaba Cloud function computing has exceeded 20 billion times a day, effectively supporting the peak of double 11 million QPS over the years, with business growth of more than 300%, and the overall scale ranking the first in China. It has experienced the tempering of ultra-complex scenarios. Serverless computing products embody Alibaba Cloud's technology precipitation and best practice experience in the cloud native domain.

In addition to the change of product form, Serverless also brings about the change of software R&D paradigm. Serverless cloud products are modular, composable, and highly available. They are the elements of building applications. By combining multiple cloud products through Serverless computing, you can quickly build applications.

Based on Serverless, the R&D method is upgraded to assembly-type R&D. Assembled R&D completely subverts the original software R&D method, greatly improves the R&D efficiency and flexibly responds to business challenges. According to the survey and statistics of authoritative institutions, assembly-type R&D can improve the efficiency of R&D by more than 50% compared with the traditional model.

Liu Xiaowei, head of large front-end technology of Deji Network Technology Co., Ltd., affirmed the cost reduction and efficiency improvement effect of Serverless. He said: "In the past year, we have been upgrading and transforming the front-end serverless. From the perspective of expenditure cost, we have to reduce it a lot. From the perspective of service, we need to divide it into several aspects. The cost of transforming a single application into serverless has increased appropriately, but once the transformation is completed, the efficiency of subsequent development will be much higher."

This is the value that Serverless brings: it really allows developers to return to the business itself, so that enterprises can do less and gain more.

The dilemma and break in the process of Serverless

For the current Serverless, the value is self-evident. Of course, Serverless will also encounter some resistance on the way to large-scale implementation. Enterprise developers still have hesitations and concerns about whether to adopt the technology. They agree that there is an invisible gap between the technology trend and the actual application implementation.

Ding Yu believes that this is a common phenomenon in the process of technological development and is in line with the law of technological development. The large-scale implementation of Serverless needs an opportunity, and this opportunity has come. At present, enterprises and developers are more active in embracing cloud computing, and cloud has become an innovative platform. At the same time, more products have cloud features, such as adaptive flexibility and free operation and maintenance. From the perspective of Serverless itself, because it is advanced enough, it does not simply replace the Internet distributed architecture, but upgrades the Internet distributed architecture, thus forming its own system, which can bring many obvious changes to enterprises.

Due to different periods, enterprises will have different choices. With the development of Serverless technology, it is believed that more enterprises will be willing to embrace Serverless. It may take 5-10 years to become the mainstream architecture like the Internet distributed architecture, but the first wave of dividends will often be received by the first, which will make enterprises more competitive.

At present, the Java language still dominates the domestic back-end development, while the cold start feature of the Java language is not friendly in function calculation. The cold start time makes it difficult for users to enjoy the technical dividend brought by FC millisecond charging, second scale expansion and other features.

Not long ago, AWS released SnapStart, a product that eliminates cold start, making cold start a hot topic again.

Cold start includes system cold start and application cold start. For system cold start, Alibaba Cloud has carried out full-stack optimization on hardware, MicroVM, operating system, language runtime, etc. Alibaba Cloud functional computing FC is the industry leader in performance indicators at the system cold start stage. The application cold start is related to the language selected by the user and the program logic itself. For example, the startup time of Java applications is usually much longer than that of Node.js, Python, Go and other languages. Alibaba Cloud has been polishing these technologies internally, but it will be more cautious in practical applications. Because SnapStart and other technologies require the customer's application to adapt SnapStart, otherwise the program correctness problem will occur. Today AWS supports this technology, and has done a lot of adaptation of basic libraries and third-party libraries. Although it simplifies the complexity of customers, it still needs customers to ensure the correctness. Therefore, Alibaba Cloud provides some solutions to ensure compatibility, hoping to reduce the user's adaptation burden as much as possible through continuous optimization.

At present, Alibaba Cloud mainly deals with the cold start problem through reservation and idleness. In addition, this year, the cloud conference function computing FC released the strategy of vCPU and memory unbinding. Customers can more finely manage the resource allocation, and the resource utilization is optimized again, and the cost will be further reduced.

In addition, vendor locking is also a major obstacle to the scale of Serverless. Alibaba Cloud also has a solution.

First of all, Alibaba Cloud has opened up Serverless Devs. Serverless Devs is an open source serverless developer platform, dedicated to providing developers with a powerful tool chain system. Thanks to the pluggable feature of the function, Serverless Devs can easily support project deployment of different cloud vendors or deploy to different cloud platforms with one click.

Secondly, the concept of Serverless architecture itself is to reshape the application research and development model. The cloud platform provides a complete Serverless product system, and the cloud platform provides a complete set of Serverless BaaS and a more standardized and process-based bridge. The more fine-grained business processing splitting of FaaS can make the business code development easier, and then make the construction of enterprise cloud mode easier.

Finally, the adoption of multi-cloud mode by enterprises does not mean that all businesses should be multi-cloud and evenly distributed on each cloud. Most of the time, the best cloud is selected based on the ability of differentiation. Only by selecting the most appropriate cloud and deep use of cloud can the advantages of cloud manufacturers be truly brought into play and better cloud resources be obtained. To sum up, how to do without lock-in, manufacturers are open and standardized, and there are no restrictions. To a certain extent, there is no lock-in.

Imagination and planning for the future of Serverless

In addition to the need for technology to pass the customs, the change of thinking is also a gap in front of developers.

From technical innovation to application and then to popularization, the biggest challenge for developers is the change of system design pattern and coding habits.

Many developers have questions about the learning of the Serverless architecture. Is it to use the original development paradigm or need to learn a new paradigm? With regard to this concern, Song Jie, CEO of Huiyou Culture, said: "Serverless will subvert the traditional development model and logic, which may be more challenging for mature developers. In terms of learning costs, for experienced programmers, the biggest obstacle to entering the cloud native world is not 'learning', but 'forgetting'."

Cloud native technology has broken the original limit, and there is no size limit for CPU, hard disk and memory. The more experienced programmers are, the more difficult it is to change their thinking. "Programmers need to constantly learn Serverless before they can overcome their inherent thinking mode. I hope Alibaba Cloud can provide more training to help you realize the transformation of thinking mode." Song Jie said that this is also his expectation for Alibaba Cloud.

In addition to providing more comprehensive and rich Serverless products and solutions, Alibaba Cloud, as the pioneer of Serverless in China, will continue to embrace open source, provide Serverless users with diversified, alternative and more flexible ways, and enhance the happiness of developers.

"In the future, Alibaba Cloud's entire product system will be serverless, and the integration between products and products will be more detailed, connecting the entire link from point to point, benefiting more enterprises and developers," Ding Yu said.