openEuler Meets AI: An Intelligent, Virtuous Cycle

By Hu Xinwei, Chairperson of the openEuler Technical Committee

I am thrilled to see that since openEuler went open source in late 2019, it has evolved into a thriving worldwide community by 2023.

?

The openEuler community has achieved remarkable progress

Our journey has been marked by a tenfold increase in our developer base, now boasting almost 4,000 active developers, and a rapid growth in our project scope with around 10 new projects each month. Our community has witnessed a significant rise in the number of software packages directly merged into openEuler, with over 30 packages being merged every day.

?

Our achievements have not gone unnoticed, with 6.1 million installations of openEuler-based OSs across various industries, including key public utilities. Our platform's proven reliability and versatility have earned us recognition and adoption in diverse scenarios. As we continue to innovate and expand our offerings, we anticipate even wider adoption of openEuler in the future. To achieve this, we have developed several innovative projects. For example, MICA enables real-time and non-real-time OSs to work together on a single multi-core SoC; DSoftBus helps edge devices to discover and communicate with each other automatically; KubeOS makes Kubernetes manage the OS directly; and Etmem project turns the hierarchical memory management into a unified memory pool.

?

The openEuler community currently has more than 400 code repositories and is still growing. In the future, more and more projects will start from openEuler and make their mark in the industry.

?

The intelligent era of diversified computing has arrived

The rapid development of AI and IoT technologies has fueled an unprecedented demand for computing power. This surge in demand has given rise to a plethora of innovative technologies, from AIoT to industrial digital transformation. However, these advancements have also posed new challenges for OSs.

?

To be an intelligent OS that supports diversified computing power for versatile scenarios, our openEuler follows the principle of "OS for AI, AI for OS" and strives to incorporate AI into every part of the OS.

?

OS for AI

With the ever-growing computing tasks, it is essential to have the suitable computing architecture for each task, like CPU for general computing, GPU or TPU for AI computing, and GPU for graphics. That's why DSA is gaining popularity.

?

However, we still face a significant challenge of low efficiency in using computing power. Currently, about half of the computing power in data centers is wasted, as the fixed approaches to allocating and managing computing power cannot keep up with the dynamic nature of modern computing.

?

The primary cause of this inefficiency is the inflexibility of current OSs, which treat diverse computing architectures as separate devices. At openEuler, we are working towards a solution that integrates these devices, allowing the OS to manage, allocate, and schedule resources from a global perspective. And we have already made some progress in these areas.

?

In general, as computing power continues to rise in demand, so does its waste. This not only presents a challenge but also an opportunity for OSs to play a crucial role in achieving carbon peak and future carbon neutrality.

?

?

AI for OS

The future OSs should be designed for AI, managed by AI, and offer AI capabilities.

?

Our developers are working on EulerCopilot, a project that leverages LLMs to make openEuler an AI-powered OS. System administrators can talk to EulerCopilot in natural language and ask it to collect, analyze, and fine-tune the system. EulerCopilot can then generate scripts, identify system bottlenecks, and change kernel and system parameters with the help of A-Tune and A-ops. This AI-driven approach enables faster performance with fewer resources, ensuring that services remain up-to-date and running smoothly.

?

We're also harnessing the power of AI in our build and testing processes. Our BiSheng Compiler, for instance, uses AI to produce faster and more compact openEuler binary files. And, we've enhanced our fuzz testing with AI-generated test models, resulting in higher quality and more comprehensive testing.

?

In short, by developing AI for OS technologies within the openEuler ecosystem, we're paving the way for an OS that truly serves users and businesses in the intelligent era.

?

openEuler for versatile scenarios

As AIoT becomes more prevalent, there is a growing demand for a versatile platform that can support various scenarios and meet the evolving needs of hardware and applications. For openEuler, this means embracing versatility. We recognize the importance of creating different instances for each scenario such as embedded, edge, server, and cloud. This approach enables technology sharing, on-demand combination, ecological compatibility, and easy connectivity, while openEuler has a unique advantage in this regard.

?

On the one hand, we collaborate closely with OpenHarmony. Together, we have successfully verified our solutions in scenarios like Power Harmony+ OS, offering a complete device-edge collaborative solution. On the other hand, openEuler is constantly refining its toolchain, empowering software components to be combined and customized as needed. This unified OS platform enables the ecosystems of cloud, edge, and embedded scenarios to work together harmoniously. With our platform, applications for embedded devices can be easily migrated to the edge and keep running. Moreover, our versatile platform simplifies connectivity and communication, making cross-domain innovation more accessible and achievable.

?

The future goals and challenges

The innovative combination of openEuler and AI has created an intelligent, virtuous cycle that takes computing to the next level. We have been tirelessly working on harnessing the power of heterogeneous computing collaboration to achieve ultra-large capacity and highly efficient computing, implementing PB-level generalized memory management to boost training efficiency by over 20%, and enhancing inference concurrency by over 50% through multi-dimensional heterogeneous computing scheduling. These achievements would not have been possible without the efforts of our dedicated developers from diverse fields, who have meticulously refined every aspect of our code.

?

We are committed to providing developers with an even more intelligent and efficient development and operations environment, seamlessly integrating OS and AI capabilities. As we continue to advance, we hope to get more support from our peers to jointly create a future-oriented openEuler OS for 2030.