Alibaba CTO Cheng Li: 2021 Double 11 will be a new starting point for Alibaba Good Technology

On November 11, Alibaba's Chief Technology Officer Cheng Li released a set of data: this year's Double 11, Alibaba's peak computing cost for the promotion dropped by 50% compared with last year; as of the 11th, the Xiaomanlu driverless car had delivered more than 1 million express parcels during the Double 11 period.

The traffic peak of Double 11 is the biggest technical challenge of Double 11 every year . A large amount of computing resources need to be added at the same time to ensure smooth operation. Through the more extreme elastic computing capabilities of cloud native, Alibaba's computing costs for the peak of Double 11 this year have dropped by 50% compared with last year. Cheng Li said that based on the application of 100% cloud native, cloud native technology has continued to improve this year, and the number of online business containers has reached one million, which has greatly optimized the transaction link and improved computing efficiency.

In the past few years, Alibaba has continued to deepen its computing power. Thanks to the full cloud migration, cloud-native technology, and full-stack self-developed technology innovation from databases to servers to chips, the efficiency of computing resource utilization has been continuously improved. This year, all of Alibaba's businesses are running on the cloud. Through cloud technology innovation, data center computing power has been more reasonably configured, such as Alibaba Cloud's self-developed Shenlong cloud server architecture and the use of its self-developed database PolarDB, which greatly improved the efficiency of server resource utilization and reduced computing costs.

Alibaba's self-developed chips have also begun to be widely used in the Double 11 battlefield. This year, Pingtou Ge's self-developed Hanguang chip is the main computing power to support the world's largest e-commerce search and recommendation tasks. Among them, 100% of the AI ​​computing power of Taobao's main search is provided by this chip. Compared with traditional GPUs, the efficiency of algorithms running on Hanguang 800 can be increased by nearly 2 times.

In addition to deepening its computing power, Alibaba has also spared no effort in its investment in green technology. This year's Double 11, green has become the core keyword, and technology plays a key role in it. In addition to cloud-native technology driving a significant drop in peak computing power costs for big promotions, the self-developed Hanguang chip has also significantly reduced energy consumption while improving algorithm efficiency. Compared with traditional GPUs, the energy consumption per unit of computing power is reduced by 58%. In addition to green computing power, the DAMO Academy's self-developed large model M6 with the world's largest scale parameters also supports Double 11 for the first time. Through low-carbon, high-computing power algorithm models, it helps AI-related businesses improve computing power and reduce carbon emissions. It has been used in more than 40 business scenarios. It is worth mentioning that M6 has achieved the industry's ultimate low-carbon and high-efficiency, with energy consumption of only 1% compared to the same parameter scale.

This year, the proportion of clean energy such as wind power and photovoltaic power in Alibaba data centers continues to increase, while liquid cooling technology is continuously promoted. Since 2015, Alibaba Cloud data centers have begun to explore the practice of liquid cooling technology. By immersing the server in a special insulating coolant to achieve heat dissipation, liquid cooling technology can reduce server energy consumption by 70%. Hangzhou Renhe Data Center has deployed the world's largest liquid cooling cluster, which has achieved a PUE (power usage efficiency) reduction of 1.09.

In the logistics link, Cainiao's green logistics technologies such as intelligent box cutting, green box return, and carbon bills have also been widely promoted and applied. By optimizing carton models and recommending reasonable packing solutions, the intelligent box cutting algorithm supports Cainiao's recommendations for packaging materials in warehouses across the industry, reducing the use of packaging materials by an average of 15%, and can "slim down" 530 million packages in Cainiao warehouses alone in a year; through carbon emission measurement and information-based inventory of all links in green logistics, Cainiao has also launched a "personal carbon reduction bill", and initiated express packaging recycling at 60,000 Cainiao stations, and tried out old packaging recycling at 10,000 stations, opening up a closed loop for the recycling of express packaging.

"Deeply delving into self-developed technology and green technology is ultimately about creating good technology that benefits users." Cheng Li said that good technology is always focused on improving user experience, reducing costs and increasing efficiency for businesses, and promoting the transformation and development of related industries.

During this year's Double 11, Taobao's immersive 3D live broadcast room created the first film-level self-broadcast solution for merchants, providing consumers with an immersive shopping experience. This technology breaks through the key technologies of cloud rendering, AI, complex light and shadow simulation, and cloud 3D asset library in previous film-level XR live broadcast shooting. It supports anchors to use only ordinary green screens to simulate live broadcast venues of thousands of square meters, nearly a thousand stage lights, and special camera equipment. Live broadcast effects have become the first film-level shooting solution for merchants to broadcast themselves with the advantages of low cost, reusability, low carbon and environmental protection.

In the field of cross-border e-commerce, multilingual translation technology has significantly improved the quality of source language product information and user experience with its remarkable ability to integrate rewriting and translation. It currently supports the internationalization of tens of millions of cross-border product information in more than ten countries for AliExpress and ICBU. Image translation technology automatically locates and identifies translation areas, automatically converts Chinese images into English images, and realizes the internationalization of millions of product image information on Lazada, helping merchants improve conversion efficiency.

In addition to consumer experience, technology has become an engine to promote industrial innovation and upgrading, from product design, production and manufacturing to the last mile of logistics and distribution. Taking the fashion industry as an example, how to efficiently and accurately mass-produce innovative design styles that meet market trends and shorten the new model cultivation cycle has always been a market pain point. The intelligent trend design solution supported by DAMO Academy's AI image generation technology has become a boost to the Double 11 quick response fashion design. Many shoe and boot merchants have launched thousands of AI-designed products based on this solution. The large model M6 has also been put into use in some factories. Its high-definition image generation capability can increase the work efficiency of designers by 5 times, and the development cycle of new clothing has been shortened from the previous 1-3 months to the fastest 2 weeks . At present, the first batch of AI-designed clothing has been launched on Taobao.

In the last mile of logistics and delivery, Xiaomanlu's driverless cars drove into more than 200 universities and communities across the country during Double 11 this year, allowing users to collect packages in a time-saving and labor-saving manner through the "on-demand appointment, scheduled door-to-door, and delivery to the building" service. As of the 11th, Xiaomanlu has delivered more than 1 million packages. In the past year, Xiaomanlu has completed the mass production ramp-up from "0 to 1", verified the technical and commercial possibilities, and has also become a cute and convenient artificial intelligence assistant in the eyes of many people.

"Double 11 is an important technological training ground. It not only refines high technologies such as cloud computing and artificial intelligence, but also drives us to continuously create good technology. Whether it is participating in social carbon reduction actions through green technological innovation, or improving user experience through human-computer interaction technology and autonomous driving technology, we are constantly moving towards the direction of good technology. This year's Double 11 will become a new starting point for Alibaba's good technology." Cheng Li said.