This is a groundbreaking and leading test!
The protagonist of the test is Inspur Yunhai OS. Its typical performance is that it has completed the world's only SPEC Cloud benchmark test for the "one cloud, multiple cores" scenario, and in the mixed deployment cluster test of three types of processor nodes, its relative scalability, Indicators such as average instance configuration time have reached the world's leading level, verifying the high efficiency, high performance and high scalability of Yunhai OS in cross-processor architecture scenarios for business applications. The significance of this test is not only to once again demonstrate the power of Inspur Yunhai OS, but also to create a cloud platform reference benchmark with "one cloud, multiple cores" as its core.
Inspur Yunhai completed the industry’s first SPEC Cloud benchmark test in a single-cloud, multi-core scenario
Although this test is a personalized behavior of one manufacturer, it reflects another evolution of the cloud computing industry. "One cloud, multiple cores" is not only a fundamental solution to solve the heterogeneity of the underlying multi-computing and ecological discreteness, but it is also one of the core capabilities necessary for current and future cloud platforms.
“One cloud, multiple cores” is an urgent need
You may have noticed that top domestic and foreign cloud service providers, including Amazon Cloud Technology and BAT, have launched self-developed chips. The most direct driving factor causing this situation is the explosion of computing power brought about by the new wave of AI. In fact, the battle for computing power caused by AI is just the tip of the iceberg. Hidden deeper "under the sea level" are the challenges posed by multiple computing to hardware architecture, software platforms and even upper-layer applications.
The formation of the multi-computing market starts with the underlying chips. As we all know, based on different chips, a variety of server machines can be built, on top of which there are operating systems, middleware, databases, and even various application software. Without exception, these software and hardware systems must adapt to the underlying chips. When the type and scale of computing power increase linearly, this is the most troublesome thing for both IT manufacturers and industry users.
As a software developer, do you have to adapt to different chips every time you develop a piece of software? As a user, should you purchase one type of chip or multiple types of chips? The result is a dilemma and no clue what to do! In the past, whether it was a private cloud or a public cloud, it was mainly adapted to one of the most mainstream chips. However, after the rise of multi-dimensional computing, adapting different chips is not only a technical choice, but also a commercial choice. Because if only one kind of chip is supported, then users will have no second choice, and the risks that will follow are inevitable.
To solve the problem of heterogeneous underlying chips, different manufacturers will approach it from different dimensions and levels. Inspur Yunhai chose to stand at a higher level, that is, the cloud operating system level, to solve this thorny problem with "one cloud, multiple cores" question. Born out of "hardware reconstruction and software definition", the cloud is completely decoupled from the underlying hardware, and one cloud is adapted to different chips. This is the essence of "one cloud, multiple cores". "For Inspur Yunhai, 'one cloud, multiple cores' is not an advanced capability, but a basic capability. As long as they purchase the Inspur Yunhai OS that supports 'one cloud, multiple cores', users do not have to consider which underlying layer is used. With this kind of chip, this cloud can calmly deal with all the business and application problems of the upper layer." said Liu Jian, director of cloud computing solutions at Inspur Data.
The above has analyzed the causes of "one cloud, multiple cores" from the perspective of architecture and application. Next, we will analyze the changes brought about by "one cloud, multiple cores" from a deeper ecological perspective.
Looking back on the development history of IT, after breaking the vertical closed ecology of small computers, the IT ecology is hierarchical. Those who make chips, those who make complete machines, and those who make operating systems. Everyone performs their own duties, which is basically rare in the industry. There is no one company that is so dominant that it can take over everything from applications to underlying chips. Considering the issues of sustainable development and security, no user should be willing to invest all their wealth in a chip that currently looks good, because once there are any problems with the evolution of the chip itself, migration will be impossible for users. Bear the weight.
Zhang Dong, Chairman of Inspur Data
"So, users hope to have a more reasonable ecological environment to provide support." Zhang Dong, chairman of Inspur Data, explained, "We cannot interpret 'one cloud, multiple cores' only from a technical level. We believe that as long as there is 'one cloud, multiple cores' "Cloud with multiple cores" can shield all technical problems faced by chips and solve them once and for all. From the perspective of Inspur Yunhai, we encourage users to use "one cloud with multiple cores" as a basic choice idea and starting point when building a cloud. Rather than placing all your bets on a certain manufacturer or a certain chip architecture. From chips to complete servers to cloud operating systems, manufacturers at every level in the entire ecosystem need to work together to achieve 'one cloud, multiple cores' The mechanism can more flexibly adapt to possible changes in the underlying chip, and even if changes occur, it can ensure smooth and stable switching without affecting business continuity."
To sum up, on the demand side, diversified technology platforms and business applications such as big data, artificial intelligence, edge computing, and the Internet of Things have increasingly different demands for computing power, resulting in more computing power supply gradually shifting from processors to Various accelerator transfers including GPU, DPU, FPGA, etc. In this case, the cloud platform must have the ability to shield differences in underlying hardware architecture and complete adaptive scheduling and unified management of resources based on the characteristics of upper-layer applications. And "one cloud, multiple cores" can just meet the diverse needs of users for computing power, and realize the coexistence of computing power to the unification of computing power.
On the supply side, due to the extremely rich types of chip architectures, industry users not only face the problem of chip selection, but also need to consider how various types and architectures of chips integrate with existing IT infrastructure. "One cloud, multiple cores" not only realizes the integration of cores and clouds, but also achieves the synergy of "platform + ecology".
It is foreseeable that "one cloud, multiple cores" will become a necessity for future cloud computing platforms.
"One cloud, multiple cores" free switching
Inspur Yunhai is the first in the industry to support "one cloud, multiple cores" and has rich experience in industrial application implementation. Since about 2018, Inspur Yunhai has helped hundreds of customers in the government, finance, energy, transportation and other industries build a "one cloud, multiple cores" industry cloud. For example, a large bank used a variety of chip servers and Inspur Cloud OS to complete the certification test of core MPP data warehouse products and implement large-scale production system applications; Inspur Cloud also helped a province build the largest chip-based server in China The most diverse provincial government cloud platform, covering nearly 2,000 servers with three processor architectures, and fully integrating basic software and hardware, cloud platforms, security systems, operation and maintenance management systems, application systems, etc., with a total of 104 businesses completed System migration and connection of 4 public application systems.
"Many of the government clouds, industry clouds, and private clouds created by Inspur Yunhai are already 'one cloud, multiple cores.'" Liu Jian said, "Through in-depth practice, we have slowly polished the 'one cloud, multiple cores' platform and stepped on Through various 'pits', we gradually accumulate experience, so that the value of 'one cloud, multiple cores' can be released to industry users to the greatest extent."
Based on years of practical experience, Inspur Yunhai has summarized the methodology and successful path to implement "one cloud, multiple cores". Qi Kaiyuan, technical director of cloud computing at Inspur Data, said: "The implementation of 'one cloud, multiple cores' requires a complete set of technical systems to support it. From the understanding of Inspur Cloud Sea, it is a scenario-driven system design that is application-oriented. From the CPU as the core to the system as the core, collaborative design of software and hardware is carried out to form a diverse and heterogeneous integration. At the same time, it is supplemented by software-defined technology to continuously improve the efficiency and energy efficiency of computing. At the same time, we must also comply with layered decoupling and openness. Standards. In other words, comprehensive layered decoupling is achieved from chip accelerator components to complete machines, to cloud operating systems, middleware, and applications. Only in this way can the closedness and discreteness of a single technology ecosystem be eliminated, thereby On the whole, the standardization and standardization of 'one cloud, multiple cores' will be formed."
The implementation of "one cloud, multiple cores" needs to go through a process from shallow to deep, layer by layer. Therefore, Inspur Yunhai proposed a “three-step” strategy.
The first step is to achieve hybrid deployment, unified management, and unified view. At this stage, the problem of "is there or not" is mainly solved. Using the cloud as the entrance, the cloud operating system realizes unified management of heterogeneous processor servers, covering domestic and foreign mainstream processors, providing a unified operating interface, operating methods, and unified monitoring and operation, etc., to solve the problem of cross-architecture deployment and collaboration of applications.
The second step is to achieve business traction, hierarchical decoupling, and architecture upgrade. This stage mainly solves the problem of "is it easy to use" and further realizes low-cost cross-architecture switching of applications. Architecture upgrades are achieved through the "layered decoupling" design of the resource layer, platform layer and application layer.
The third step is to establish computing power standards and full-stack multi-core to promote ecological co-construction. The ultimate goal of "one cloud, multiple cores" is to achieve complete decoupling of applications and chip architecture. This requires collaboration between upstream and downstream industry chains such as hardware, cloud, and applications, and the improvement of quantitative methods for effective computing power under heterogeneous architectures to support equivalent switching of applications between processors of different architectures. On this basis, promote the "one cloud, multiple cores" combination of applications and cloud infrastructure to create a vertically integrated solution to achieve non-aware switching across architectures at the application level. More importantly, we should unite various forces in the industry to promote the formation of "one cloud, multiple cores" industry standards.
Inspur Yunhai’s “one cloud, multiple cores” three-step technical route
Qi Kaiyuan said: "We have helped many industry users cross the first stage and realize the large-scale deployment and application of 'one cloud, multiple cores' in their businesses. Currently, we have begun the technical layout and development for the second stage. Explore and combine native technology with middleware and other manufacturers to achieve data synchronization at the platform layer, including traffic segmentation at the application layer. We hope to form some best practices and guidance programs to help more application manufacturers follow the 'one cloud, multiple Core’s strategy goes to the cloud.”
"One cloud, multiple cores" is a buzzword, especially among cloud infrastructure vendors. Although "one cloud, multiple cores" has the same name, due to differences in understanding and technical capabilities, various manufacturers have different effects when implementing "one cloud, multiple cores".
Zhang Dong stressed: "From the perspective of Inspur Cloud Sea, we hope that through the cloud, more chips with different processor architectures can coexist in a cloud, and ultimately enable upper-layer services to switch between different processor architectures at low cost or Talk about free switching. This should also be the direction the entire industry works together."
Inspur Yunhai hopes that more manufacturers can join the trend of "one cloud, multiple cores" and truly think about issues from the perspective of "one cloud, multiple cores" whether in the product development stage or in the cloud construction process. Ultimately, Construct relevant standards and specifications to promote ecological construction and prosperity.
Left-hand ecological right-hand standard
Although there are still many technical problems to be overcome in the implementation of "one cloud, multiple cores", from a longer-term development perspective, it is very necessary and important to involve the entire upper-level ecosystem. For example, databases now basically run on the same chip architecture. One thing Inspur Yunhai wants to do now is to promote the "separation" of reading and writing in distributed cluster databases, so that they can run on chips of different architectures. Inspur Yunhai is jointly working with Hangao and other database manufacturers to tackle key ecological issues.
In addition, the standard issues mentioned repeatedly above are also a key factor affecting the long-term development of “one cloud, multiple cores”. When industry development truly enters the third stage, people will no longer care about underlying hardware issues, and standards issues as an important driving force will be further highlighted. Therefore, Inspur Yunhai has been vigorously promoting the standardization of computing power. As a standardized service, computing power has nothing to do with the underlying software and hardware. It can be delivered on top of a container or through an API. At this stage, people's focus will completely shift from the bottom layer to the business layer above. Therefore, a series of standards are definitely needed to regulate this ecology. At present, relevant domestic institutions are promoting the preparation of "one cloud, multiple cores" standards. The standardization of computing power is worth looking forward to!
Using data to empower intelligence on the cloud brings about diversified application needs, thereby giving rise to diversified computing power. This is not a "characteristic" of China, but a global trend. However, in the context of the current restructuring of the domestic industrial chain, the discussion of "one cloud, multiple cores" is more necessary and of practical value.
"We have communicated with some important customers, and they talked about a key issue, which is that the business cannot be interrupted at any time. Therefore, free and senseless switching on multi-chip platforms is a 'rigid necessity' and cannot be avoided. Today, many users have already I agree with the idea of 'one cloud, multiple cores'." Zhang Dong pointed out, "With the continuous evolution of IT, as long as there is an environment that requires multiple processors, there will be a need for 'one cloud, multiple cores'. This market has huge potential. .”
(Reprinted from Yunshuzhi Observation official account)