As is well known, only powerful cloud computing can nurture powerful AI Big models, and the foundation of cloud computing is servers. How to manage these servers well? The operating system is the lowest and most critical platform software. Caught in the upheaval of the industry and the times, IT Industry practitioners need a stable community that can provide long-term support and AI Native server OS， Compete for the upcoming 'AI Decade Plan'.

As Chen Chun, an academician of the CAE Member and a representative of the dragon lizard senior advisory group, said, "The scheduling and flexibility of cloud computing, as well as the training and reasoning of large models, are inseparable from a stable, secure and efficient server operating system".

On August 30th, at the 2nd Dragon Lizard Operating System Conference, the domestic open-source operating system root community Dragon Lizard released the official official version Anolis OS 23 ， Can better support model training and AI Application, fully compatible with mainstream domestic and international applications CPU、GPU framework.

The most surprising thing is the dragon lizard OS Currently, there are over 8 million installed units. This means that the dragon lizard OS In the current implementation of over 1000 partners and 1 million users, a long-term self circulating ecosystem has been initially achieved, making it one of the largest and most comprehensive server operating systems in China.

And this is just the beginning.

refer to PC Terminal Windows The experience of the system ultimately dominating the world with the help of Intel and numerous developers: the success of an operating system is actually the result of the cooperation and co creation of the entire upstream and downstream industries.

Looking towards the future, Dragon Lizard has officially launched the "Anolis OS 23 Ecological Derivatives" “CentOS The three major plans of "substitution" and "AI application promotion" are focusing on the underlying ecology, meeting market demand, and AI Apply three levels to fully catch up.

As one of the three major obstacles in the software industry, China's local server system is entering a new stage, AI In the opportunity of large models, there is even the possibility of overtaking on curves.

AI What does a native operating system look like?

“Android Father Andy Rubin once observed the phenomenon of operating systems changing every 12 years and mentioned it in 2017, when the internet was still thriving during the era of mobile internet, “AI It is the next important operating system.

His argument still needs time to test, but in a new round AI With the rapid development of technology, mobile phones PC After all kinds of hardware, they began to have the so-called AI Native operating system to better support AI Reasoning and application.

However, in undertaking the most critical AI The server-side for large model training tasks, used for training large models AI The operating system has yet to arrive.

In the view of Ma Tao, Vice President of Alibaba Cloud Basic Software Department and Chairman of Dragon Lizard Community, the main reason behind this is that different operating systems on different ends have completely different scenarios and difficulties to face and handle.

Mobile phone AI， Perhaps the operating system supports wake-up Siri； Desktop operating systems, such as windows， Maybe it's support AI Create a schedule, write a summary, and other tasks. But the server operating system is completely different, and large model training now mostly runs on the cloud, which poses greater challenges for server operating systems that schedule and manage computing resources. On the other hand, an operating system running on a cluster of tens of thousands or hundreds of thousands of servers requires AI Analyze the system's difficulties and risks. ”

The feelings of server hardware manufacturers are more obvious. Zhang Dong, Chief Scientist of Inspur Cloud Sea and Vice Chairman of Dragon Lizard Community, bluntly stated, “ AI Technology has developed too rapidly in the past two years, and the underlying hardware and operating systems have been pulled away. ”

The user said that stuffing 8 cards into a server is not enough, but keeping 16 cards is not. Storing 60 disks is not enough, and you have to match 100 disks, which will soon be 200 yuan. This forces hardware manufacturers to make the machines bigger and bigger. The cluster size is also getting bigger and bigger, and 100 cards are not enough. We need 1000 or 10000 cards. How to efficiently manage and schedule these hardware resources? The operating system is a very important link to solve this problem

On the one hand, the operating system itself needs to be able to handle AI The explosive growth of related hardware and compatibility issues with heterogeneous hardware; On the other hand, it is necessary to use AI The ability to transform the operating system, automatically handling complex tasks such as adaptation, setting up environments and systems, making it easy for users to use directly.

I think we should move the operating system towards AI The future direction of development is certain, but now AI The adaptation is actually far from enough. How to further transform server operating systems into AI Native, better supported AI Training and reasoning, while the operating system itself can also become an intelligent agent, require greater levels of innovation. ”Zhang Dong summarized as follows.

Dragon Lizard is trying to solve this problem, with "System for AI" on one side and "AI for System" on the other.

Specifically, “System for AI” The main reason is that the system has conducted extensive optimization work on the compatibility, stability, and security of training and inference for large models, in order to better support them AI development.

Newly released Anolis OS 23 Official version, using ANCK 6.6  The kernel significantly enhances compatibility with multiple platforms and fully supports mainstream domestic and international applications CPU、GPU framework. Regarding AI Widely used in scenarios AI In the framework, the following are provided: OpenVino Native support included.

And, Anolis OS 23  Adapt to updates, richer, and more secure AI  Alibaba Cloud AI Containers, including AI on NVIDIA、 AI on AMD、AI on Intel  and AI on  domestic GPU  Waiting for multiple ecological scenarios.

Container services currently account for 80% of cloud based services AI Tasks are the most mainstream AI The development method, this iteration of the new version of Dragon Saurison, is bound to help more AI Reasoning and application grew directly from the Dragon Lizard operating system.

On the "AI for System" side, the main consideration is the efficiency and usability of users in using Dragon Lizard, which has been strengthened AI The advantages of native operating systems. The Dragon Lizard operating system was created using a large model AI assistant Copilot， Capable of answering user questions, performing simple operations, and analyzing system issues.

In addition, dragon lizards also explore and utilize AI Ability to assist system administrators, R&D personnel, security and operations personnel in better using this operating system, users will feel that it is based on AI The design truly embodies the meaning of 'AI native'.

A good operating system requires collaboration across the entire software and hardware industry chain

The operating system is composed of tens of thousands of software packages, which are like tens of thousands of cats on the street. The operating system needs to make tens of thousands of cats line up in a short time W Form and arrange in a timely manner S The form and difficulty can be imagined. ”In the opinion of Cui Zhan, the general manager of Tongxin Software Server Product Line, making a good operating system is not easy.

Even more difficult is to create a widely used and successful operating system. Throughout the entire process IT The development history of the industry heavily relies on the joint efforts of the upstream and downstream of the industrial chain.

exist PC During this period, it was Microsoft Windows In the early days, he firmly held onto Intel. Perhaps, Intel's X86 Architecture is not necessarily everything CPU The most optimal instruction set, from DOS System iteration Windows It may not necessarily be the best either PC Operating system. But in PC During the early period of infiltration, “Wintel” The alliance relies on integrated software and hardware cooperation to PC The first batch of programmers in the industry co created Windows give Intel The global dominant position.

In the era of mobile Internet, this cooperation has become ARM。 Android pursues cheaper and more customized chip hardware, ARM Architecture was just the best choice at that time. The two sides joined hands to forge the mobile Internet era AA Legend (Android&ARM).

through Windows and Android The successful experience shows that in order for an operating system to succeed, in addition to having sufficient performance, it also requires collaborative innovation across the entire industry chain from hardware to software.

As a founding member of the Dragon Lizard community, Alibaba Cloud proposed the concept of "one cloud, multiple chips" in the past two years, with chips from different manufacturers and functions at the bottom and a unified cloud that outputs computing power at the top.

To achieve this goal, it is necessary to achieve maximum compatibility at the critical platform software layer of the server operating system.

Newly released by Dragon Saurison Anolis OS 23 The official version significantly enhances compatibility with multiple platforms and updates development tools and languages GCC We have made special optimizations for domestic chip platforms, which can bring an 11% performance improvement.

Dragon Lizard is fully compatible with domestic chips and can also provide good support for international mainstream chips.

Intel is also one of the governing units of the Dragon Lizard Community. Yang Jiguo, Senior Technical Director of Intel and Vice Chairman of the Dragon Lizard Community, proposed that "enterprises should CentOS After transitioning to Dragon Lizard, there will be no obstacles in terms of performance and compatibility.

On the one hand, Intel's latest chip products are also compatible with Dragon Lizard, such as Anolis OS23 It was the first to support Intel's recently released Xeon 6 chip platform this year; On the other hand, for widely used Intel chips, Intel can continue to provide compatibility and ecological expansion support in the Dragon Lizard community.

From CentOS Moving to Dragon Lizard, we found that Dragon Lizard may do better, faster, and more efficiently in terms of support for the new platform and chip optimization. Yang Jiguo said.

Yang Jiguo also revealed that Intel has done a lot of work in the Dragon Lizard community, making the Dragon Lizard operating system compatible with AI Hardware can be better compatible; At the software framework level, Intel has integrated open heterogeneous programming frameworks into the Dragon Lizard community, allowing users to do so in a very open and open-source manner AI Development work.

Another giant in the field of chips Arm， We are also exploring how to better contribute to the Dragon Lizard community.

At this year's Dragon Lizard Conference, Arm、 Alibaba Cloud, Pingtouge, ZTE New Focus and other companies have also jointly announced the establishment of the Dragon Lizard Community Arm Working group, collaborative promotion based on Arm The basic software ecosystem of architecture.

The bridging role of the operating system is amplified through the collaboration of the open source community, enhancing the effectiveness of the system. Through the efforts of all parties, the Dragon Lizard Community has now gathered over 1000 community participants and partners, making it one of the largest and most comprehensive operating system root communities in China.

This will obviously benefit every member of the open source community as well.

Jiang Jiangwei, General Manager of Alibaba Cloud Infrastructure Business Unit, bluntly stated that thanks to the active participation and contribution of numerous manufacturers of general-purpose heterogeneous chips, especially domestic self-developed chip manufacturers, in the Dragon Lizard community, Alibaba Cloud can better develop its one cloud multi-core strategy. While obtaining a more robust hardware supply chain guarantee, it also realizes unified resource management and scheduling, thereby providing more efficient computing infrastructure services to customers.

Unified kernel, adhere to open source, solve fragmentation problems

Data shows that in 2023, China's platform software market will grow rapidly, with a scale of 81.66 billion yuan, a year-on-year increase of 17.4%. The growth rate of China's operating system market has further accelerated, reaching 23.2%, and the main driving force for the growth of the operating system market comes from server operating systems.

The rapid development of the operating system market is accompanied by the troubles of inconsistent underlying kernels and fragmented versions.

Zhang Dong frankly stated, "There have been many versions of operating systems, and the situation in China is also quite complex, possibly even more complicated than abroad. As a whole machine manufacturer, the fragmentation problem we faced in the past application promotion process is a headache for us. Because any of our devices must undergo extensive testing before leaving the factory, and every new component introduced must be tested. During the testing process, mainstream operating systems on the market must be run once

The Dragon Lizard Community has proposed a new plan for this.

We hope to achieve this through Anolis OS23， Confirm many compatibility issues through standards, specifications, and other means to form a relatively unified and stable foundation. For example, hardware manufacturers only need to adapt Anolis OS23， In theory, it can be adapted to any product based on Anolis OS23 The commercial version, such as the 12 derivative versions currently available, can be adapted to reduce the upstream and downstream costs of the entire operating system ecosystem. ”Ma Tao explained that this is Anolis OS23 The most important significance of ecological derivative plans.

Anolis OS 23 The ecological derivative plan requires the integration of technology core, supply chain and other community participation standards, and the release of corresponding commercial derivative versions, community open source versions and other different versions. In this way, the entire software ecosystem in China and the future business upstream and downstream can have a unified mechanism for the kernel, toolchain, and KAPI， Furthermore, it will promote the ecological development of domestically produced operating systems in China.

At the Dragon Lizard Conference, Academician Wang Huaimin of the Chinese Academy of Sciences specifically mentioned that under the coordination of national departments, Chinese open-source operating system communities such as the Dragon Lizard Community have already Linux Consensus has been reached on the selection of kernel versions and related runtime packages.

get rid of Anolis OS 23 In addition to the Ecological Derivative Plan, the Dragon Lizard Community has also launched two major plans: the "CentOS Replacement Plan" and the "AI Application Promotion Plan".

CentOS On June 30th of this year, the service was completely shut down, and many enterprises are facing challenges of migration and continuity. The Dragon Lizard Community has done a lot of work APP、 Adaptation of software ecology, hoping to achieve through Anolis OS23、 By utilizing various version upgrade and migration tools, users can easily migrate to Dragon Lizard.

In Cui Zhan's opinion, the Dragon Lizard community is very responsible: "CentOS server shutdown will lead to business shutdown, and it takes time for users to completely detach from this platform after the shutdown. The Dragon Lizard community has set up a group specifically for this purpose CentOS The operation and maintenance supervision, with the participation of Tongxin Software, has provided extensive support for patch maintenance and upgrades. ”

“AI The 'Application Promotion Plan' represents the future. At this conference, the Dragon Lizard Community launched its first "AI native operating system" development roadmap, focusing on AI The era has also launched AI Container mirroring, intelligent operation and maintenance AIOps、OS Copilot Three major plans for document construction, continuously promoting the development of the Dragon Lizard operating system Sys for AI and AI for Sys Continuous breakthroughs in two directions, reshaping the operating system AI The core competitiveness of the times.

Ma Tao summarized: "The three major plans, in simple terms, are that we Anolis OS23 As the core, we will promote the development of the open source ecosystem of operating systems through the use of the Dragon Lizard operating system as the core, focusing on both 'continuation' and 'opening up' aspects. ”

The open and open-source ecosystem is the keyword of the Dragon Lizard community, which has also become one of the reasons why many top companies are attracted to participate in Dragon Lizard.

Yang Jiguo candidly admitted that he has been doing open source for more than 20 years, and there is not much difference between China and the international community in terms of technology and philosophy of open source. "People who do open source identify with this concept: an open mindset and an open development model jointly promote technological development

He also observed that unlike foreign open source communities led by commercial companies with commercial purposes, China's open source community is more like a real community, where everyone has the same goal and participates together to contribute.

Like the Dragon Lizard community, we adhere to openness, neutrality, and are a one person, one vote community, so basically this community can have a better mechanism to bring together common business partners, including Intel and Intel's competitors, in the community. From a technological development perspective, it can promote the development of open source communities, which is a better model. Intel is also very willing to invest in open source communities like Dragon Lizard, "said Yang Jiguo.