Today, the Alibaba Cloud Summit Guangdong-Hong Kong-Macao Greater Bay Area opened in Guangzhou. Ding Yu, a researcher at Alibaba and general manager of Alibaba Cloud’s intelligent cloud-native application platform, spoke at the forum. He said:
Serverless leads a new development paradigm on the cloud. Through rich atomic services, full hosting, high flexibility, and O&M-free advantages, out-of-the-box scenario-based capabilities, and a more cost-effective pay-as-you-go model, it helps enterprises leapfrog technology gap, making innovation within reach.
In the past ten years, going to the cloud has become a deterministic trend. In the cloud migration stage, enterprises focus on how to achieve smooth cloud migration, so cloud vendors regard cloud hosting as their core strategy. 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 has changed to how to make better use of the cloud's capabilities to quickly push products to the market, thereby realizing business success.
However, if computing power is still presented in the form of resources such as servers, the threshold for its use is still very high. Computing power and business are too far apart, and enterprises need to have a complete set of infrastructure supporting applications to make good use of computing power. To make computing power as popular as electricity, cloud computing needs a new form, which is Serverless.
Development paradigms will be redefined as cloud use becomes key. Through product service and full hosting, enterprises and developers can focus on business logic development; and cloud services have the advantages of orchestration and reusability, allowing enterprises to do less and gain more; based on Serverless, it can easily Build highly elastic applications, allowing enterprises to calmly deal with traffic fluctuations. Under the serverless development paradigm, the delivery cycle of new enterprise functions is greatly shortened, which further accelerates business iteration and wins market opportunities.
Alibaba Cloud Serverless Container Service ASK New Upgrade
As an important technical component of cloud native, K8s has been widely recognized by developers and enterprises, but its own complexity and steep learning curve are still daunting.
Alibaba Cloud released the first serverless container service ASK in 2018. Its essence is to decouple the runtime of the container from the specific node operating environment, allowing users to directly deploy applications without managing K8s nodes and servers, greatly reducing the cost of container services. usage threshold.
The new upgrade of ASK further integrates the capabilities of Alibaba Cloud infrastructure, bringing significant improvements in usage costs, creation efficiency, heterogeneous resource compatibility, and flexible supply guarantees, and solves the complex challenges developers face when using K8s. It also complies with the new demands arising from the explosion of AI scenarios.
Fully managed components, zero K8s operation and maintenance costs: ASK has newly added hosting support for more than ten K8s core components such as CoreDNS and Metrics Server, providing dynamic capacity planning capabilities, customers do not need to deploy and maintain themselves, and minimize the complexity of operation and maintenance. At the same time, ASK has also added intelligent risk identification capabilities to support automatic upgrades of K8s versions to avoid application failures or abnormal risks caused by upgrades.
Extreme elasticity, accuracy rate increased by 80%: ASK provides the world's first container image cache-based AI large image pull capability in seconds, and AI application startup time is reduced by 90%. It also provides end-to-end elastic acceleration, full-stack optimization for AI/big data workload containerization, and improves access performance by 30% through data set acceleration; in addition, this ASK also enhances the ability of intelligent elastic prediction AHPA, compared with manual configuration , the elastic accuracy rate is increased by 80%; at the same time, new support for GPU is added.
Inclusive computing power, 40% price reduction: In order to provide customers with better services, release technical dividends, and make computing power more inclusive, ASK has added support for U instance specifications, and supports multiple processors in a unified manner. Instance price reductions of up to 40%. Added the SavingPlan elastic version, which is oriented to the application of non-fixed peak and trough scenarios, and the additional cost is optimized by more than 10%. In order to further make prices more transparent, support for cost suites has been added to provide a clear insight into the cost of elastic resources and make cost management more convenient.
"Shuhe Technology adopts Alibaba Cloud ASK to deploy online models without K8s node management, dynamically uses POD according to real-time traffic, and saves resource costs by 60%; through the ASK Knative service, it solves the problem of gray scale release and multi-version coexistence of Shuhe models ; Thanks to the advantages of ASK's automatic scaling and shrinking to 0, operating costs are reduced and service availability is greatly improved." Ding Yu introduced.
Function Compute FC: Make AIGC Application Development Easier
In 2023, generative AI will usher in a concentrated explosion, and at the same time, the demand for GPUs will not increase. Alibaba Cloud Function Compute has extremely flexible GPU instances and large-scale Function Compute performance instances, which are important links for carrying stable and high-performance reasoning of AI applications. At this summit, function computing GPU ushered in an upgraded performance experience:
More flexible user configuration: Function Compute provides the industry's smallest GPU virtualization granularity, with a memory specification as small as 1GB. The CPU and GPU are decoupled, allowing users to configure them independently, and support GPU Turing/Ampere two generations of cards at the same time.
Higher resource utilization: The underlying technical architecture has transitioned from the ECS architecture to the Dragon GPU architecture, and the industry's first multi-tenant secure GPU sharing virtualization solution has increased the overall resource utilization by 80%. Finely match the AI reasoning load type, and the minimum computing power specification reaches 1/16 T4, 1/24 A10.
The technology is more advanced: the function computing GPU cold start time is from minutes to seconds , and the performance is improved by 300%. The industry's first GPU pay-as-you-go , cut into quasi-real-time inference scenarios, support AIGC popular scenarios, and help AI entrepreneurship and productivity improvement.
We found that in practical applications, the threshold of AI application development technology is still very high for many people; in addition, deploying AI applications in production environments needs to consider issues such as security, reliability, scalability, and maintainability. Certain technical skills and experience are required.
Function Compute is committed to providing AI developers and enterprises with high-performance, low-cost AI application development and deployment services. The serverless architecture has the characteristics of high resource utilization, pay-as-you-go mode, and free operation and maintenance on the server side, allowing developers to truly develop AIGC applications with zero technical threshold.
Integrate complete application cloud suites such as Serverless Application Center and Serverless Devs: Help developers complete business development from 0 to 1 and then to N, and provide full application life cycle management capabilities. Through the Serverless Application Center, users can quickly deploy and manage applications without performing additional cloning, building, packaging, and publishing operations before deploying applications, and easily accumulate best practices.
Build a complete AIGC competence center: Alibaba Cloud products are highly integrated with AI frameworks such as LangChain, and developers can develop and deploy models in open source ecosystems such as ModelScope and HuggingFace or community selection models.
Based on the function computing + serverless application center, developers can one-click model hosting, get started with AI application development in 5 minutes, and improve R&D efficiency by 80%.
The follow-up Serverless Application Center will continue to accumulate typical AI application case templates from various industries, so that users can understand and master it more easily. At present, the Serverless Application Center has access to more than 10 popular AI application templates such as Tongyi Qianwen, Wenshengtu, Tushengtu, and Tushengwen.
Function Compute opens a new window for the application of AIGC, "so that everyone can develop AIGC applications." Ding Yu introduced .
Based on Function Compute's FC+Serverless application center capabilities, Alibaba Cloud launched a new "Function Compute One-click Deployment Tongyi Qianwen Pre-Experience" , becoming the industry's first application platform that can try Tongyi Qianwen, combined with business scenarios, and successfully deployed Tongyi Qianwen pre-experience application can get 30 dialogue opportunities.
In addition, classic AI scene experience activities such as Wenshengtu, Tushengtu, Tushengwen, and Wenshengwen were also launched this time, allowing developers to complete AIGC application deployment in 5 minutes, allowing ideas to be implemented faster.
Free trial and upgrade of cloud native products
Previously, Alibaba Cloud released the "Feitian Free Trial Plan", which provides free trials of 50 full-stack cloud products, including function computing, ECS, database PolarDB, machine learning PAI, etc., for tens of millions of cloud developers in China. Serverless development model.
This free trial and upgrade of cloud-native products not only adds serverless container service ASK, but also will soon launch a variety of products such as cloud message queue MQ, serverless application engine SAE, performance testing, etc., to further enrich the usage scenarios of enterprises and developers.
In addition to the free trial plan, Alibaba Cloud has also built rich content such as cloud native communities, developer training camps, training systems, and experience scenarios, allowing developers to build the desired architecture through multiple free trial products with one click. , quickly experience the charm of cloud native and serverless.
Ding Yu said that Serverless is committed to making computing power more inclusive, allowing more people to enjoy the technological dividends, allowing innovation to flow, and enabling everyone to become a new developer in the cloud-native era.