Intelligence, high efficiency and new interaction will reshape the Internet

01 The new era puts forward higher requirements for computing infrastructure

The Internet is evolving into a new era of intelligence, efficiency and interaction, which puts forward higher requirements for computing infrastructure.

Enterprises want to provide users with a more extreme experience, which needs to rely on data intelligence technologies such as intelligent recommendation and big data mining to carry out more efficient data processing and call for computing infrastructure with lower latency and higher performance; At the same time, the iterative strategy of "running in small steps and trying to make mistakes quickly" requires agile and efficient infrastructure to support low-cost and rapid business trial and error; Facing the future, immersive experience, XR, etc. have begun to rise, and the Internet has shown the trend of 2D gradually becoming 3D, which requires the support of large-scale computing power and full-platform visual computing ecosystem.

Alibaba Cloud elastic computing, from the aspects of infrastructure, self-service capability and ecology, comprehensively builds a cloud infrastructure for Internet enterprises to help enterprises meet the new era.

02 Re-upgrade elastic computing ECS to ensure smooth and stable experience

First of all, in the underlying infrastructure part, as enterprise applications continue to put forward requirements for strong computing power, low latency, and high bandwidth, the traditional CPU-centric architecture faces more and more challenges. Alibaba Cloud has made systematic innovation from the internal architecture of the data center, changing the previous CPU-centric architecture into the cloud operating system+CIPU-centric architecture.

With the support of Apsara operating system+CIPU, Alibaba Cloud is about to launch the eighth generation of elastic computing instances. Its data processing efficiency will be greatly improved, which will greatly help customers to reduce costs and increase efficiency.

In terms of computing power, the latest processors are selected for the eight-generation main selling examples, including Intel's latest fourth-generation Xeon scalable processors. Based on the improvement of general computing power, it has the ability to support hardware acceleration natively. In the field of AI, the performance of training scenarios has been improved more than twice; The performance of encryption and decryption, compression/decompression and other scenarios is improved by more than 4 times, providing customers with scenario-based ultra-high cost performance.

At the same time, ECS instances continue to build ultra-high security water levels, fully equipped with TPM trusted chips, to ensure the integrity of the cloud operating environment, and to achieve the security of native computing on the cloud in combination with encryption computing capability and memory encryption TME technology.

In addition to the improvement of computing power and security capability, the network and storage I/O have achieved significant evolution, which is compatible with NVMe cloud disks, supports sharing, and reduces the latency to 100 microseconds; The physical network is upgraded to 2 * 100G super bandwidth, and the self-developed eRDMA acceleration capability is fully equipped as standard, significantly improving the network transmission efficiency.

When it comes to eRDMA, people subconsciously think that it is only for high-performance computing applications, but it is not. ERDMA is compatible with the standard network interface, without changing any business logic. Most applications can get performance improvement as long as they adapt to the interface.

The minimum delay of TCP physical network is still more than ten microseconds, while the minimum delay of eRDMA is 5.5 microseconds, which brings 80%, 30%, 30% and 38% performance improvements to Redis database scenario, Spark big data scenario, AI training scenario, HPC scenario and other scenarios.

The Yitian 710 chip released at the Yunqi Conference last year has been tested very well in internal and external customers of the group, and will be fully commercialized on the 15th of this month.

It is worth noting that the architecture of the ECS instance is completely different from the traditional computing architecture. In the past, the dual-channel server needs dual-CPU connection, and now the CPU becomes the peripheral of the CIPU. The CIPU can not only connect to Yitian 710, but also connect to other chips, and get better performance.

With the use of Yitian 710, compared with traditional examples, the cost performance of the codec scenario has been improved by 80%. The database scenario, AI reasoning scenario, and big data scenario have also been effectively improved by 30%, 70%, and 50%, respectively. We believe that through the continuous development and progress of technology, we can make our products more cost-effective and reduce costs and increase efficiency for customers.

03 FinOps x CloudOps, efficient support for agile development

In addition to selecting examples with high performance and high cost performance, how to better adapt the customer's business to the cloud and how to find the right "posture" for using the cloud are the two primary driving forces for reducing costs and increasing efficiency.

For example, as a global technology service platform, Huiliang Technology has built a highly elastic cloud native business architecture. After using the savings plan+pay-as-you-go model, the cost on the cloud has decreased by 20%.

In addition to flexible combination of payment modes, reasonable use of CloudOps cloud automation O&M suite can also help enterprises develop O&M engineers to improve efficiency while reducing costs. Some cloud enterprises are faced with high cost of elastic capacity management, low efficiency of application release and deployment, and slow positioning of service exceptions. After using CloudOps, the cost is reduced by 30%, the efficiency is increased by 50%, and the MTTR is reduced by 35%. CloudOps suite can bring efficient and intelligent cloud experience, and bring more possibilities of cost reduction and efficiency increase.

04 Visual computing platform to support the exploration of the next generation of Internet

Innovation is the eternal theme of Internet enterprises. Visual computing is considered to be the key technology of the next generation of Internet. On the basis of IaaS capabilities, we need to work with more ISV partners to build a visual computing ecosystem.

As an important part of the ecosystem, Alibaba Cloud and Epic Games, the parent company of the well-known 3D engine Unreal Engine in the industry, have fully cooperated. It is believed that it can effectively reduce the development threshold of 3D applications and accelerate enterprises to explore the next generation of Internet.

Xu Liang'an, the enterprise-level business director of Epic Games in Greater China, was invited to share the experience of cooperation with Alibaba Cloud:

"After more than 30 years of efforts, we have also accumulated many technologies in the field of graphics rendering, including the" nuclear weapons "in the fifth generation of virtual engine: real-time ray tracing and infinite polygon technology; Through Epic Games Online Service, developers around the world can easily connect users across platforms, countries and industries. In order to welcome the advent of the web 3.0 yuan universe, we will also open our 3D asset platform to you for free, benefiting the entire industry.

Open source is the concept that Epic Games first emphasizes, and the equally important business concept is inclusive. Fantasy engines are free in most cases. In a small number of cases, customers can make money first and then make progress together.

We are convinced that the web3.0 era is 3D, HD and interactive, so the virtual engine is indispensable; At the same time, web3.0 will also be cloud-based, so I am very happy to work with Alibaba Cloud to help developers and business partners more efficient, so that users can gain a better experience.

”

Xu Liang'an, Enterprise Business Director of Epic Games in Greater China

05 "Feitian Acceleration Program" will increase subsidies of 10 billion yuan to accelerate enterprise innovation

Alibaba Cloud elastic computing focuses on the training of talents in the cloud computing industry while building product technology. Through the smooth implementation of the Apsara Acceleration Program, we can see that hundreds of thousands of teachers and students from more than 2200 colleges and universities have participated in the Apsara Acceleration Program, and can experience up to 1.16 billion computing hours on the cloud for free.

This year, the Apsara Acceleration Plan will be upgraded to phase 3.0, and subsidies will be added to small and medium-sized enterprises. The subsidy covers 47 cloud server specifications, with a maximum subsidy period of 3 years and a maximum margin of 64%.

Today, the cost of offline use of data centers by small and medium-sized enterprises is still very high. I hope that through the Feitian Acceleration Plan, we can grow together with small and medium-sized enterprises on the cloud. I also hope that with the support of Alibaba Cloud's elastic computing product capability and cloud native computing infrastructure, we can better escort enterprises to achieve innovation.