Alibaba Cloud CIPU is a “native” adventure

In the past few decades, behind many business changes, if you go back, you can find the starting point in the changes in computing. From the continuous rise of computing power following Moore’s Law, which eventually unlocked smartphones and brought the mobile Internet; to the emergence of cloud computing, which solved the economic problems of ultra-high concurrency and large computing power; and now pursues the credibility and security of computing , rather than the blockchain of computational efficiency and computational economy… It can be said that the core point of “computing” is the eternal development foundation of the digital economy. Changes in technological capabilities, and in the goals it aims to solve, each tick means a business change.

Of course, no matter when, the improvement of computing efficiency is an important issue, and the improvement of efficiency will eventually return to the bottom layer of chips and systems to find space.

Recently, I saw that Alibaba Cloud did one thing, released a self-developed cloud data center dedicated processor CIPU (Cloud infrastructure Processing Units), and said that the CIPU is a special-purpose processor designed for the “Flying” operating system. The processor will replace the CPU and become the processing center of the new generation of cloud computing architecture.

What puzzles me here is that new concepts are not scarce in the cloud computing industry, and it is already overwhelming to include IPU, TPU, DPU and various AI chips. And what exactly is the CIPU launched by Alibaba Cloud? How is it different from the previous concept? What will it bring us?

01 The plight of CPU, the origin of CIPU

I think Alibaba Cloud’s launch of CIPU is actually a “long-planned plan”:

The “Feitian” they launched before is said to be the only self-developed cloud operating system in China, and they have been investing in self-developed cloud technology over the years; at the Yunqi Conference last year, Panjiu, Yitian, Shenlong 4.0 and several blockbuster products of Lingjie are “building a deep foundation” and “growing downwards” step by step.

get?code=MDdmYzZmNjM3NmU4ODQ2MDE5YzcxOWU Figure|Alibaba Cloud CIPU released

This time, the CIPU is specially designed for cloud data centers, which can still be regarded as a continuation of Alibaba Cloud’s “downward growth” logic.

But to better understand the emergence of the CIPU, it is necessary to talk about the changes in the data center over the years.

There is an ancient term in the data center called “computer room”, which sounds a bit old and industrial. In the early stage of Internet development, it was indeed not high-tech, and it was more used for stacking computers. However, with the maturity of the Internet, data centers have increasingly become “the gathering place of business and traffic complexity”, and a series of technological innovations have emerged.

A number of technology giants such as Amazon, Microsoft, and Google have also sensed opportunities in the data center field, and have invested heavily in self-research and self-construction. Alibaba Cloud announced in April 2020 that it will invest another 200 billion yuan in the next three years for the research and development of major core technologies such as cloud operating systems, servers, chips, and networks, as well as future-oriented data center construction.

In addition to seeing opportunities, the giants spend so much effort in self-developed data centers, but also because the software and hardware of traditional data centers are still lagging behind in terms of performance and cost, and cloud data centers can solve large-scale computing problems. Since the existing facilities cannot be satisfied, we should simply re -design the cloud environment for a more native design, and maybe even systematically reconstruct it. For example, CPU chips are not actually designed to be equipped with cloud operating systems.

A classmate of Alibaba Cloud told me that in the past ten years, the development of cloud computing technology has gone through two stages: the first stage is distributed technology, which promotes the entire migration of Internet companies from mainframes to distributed systems; the second stage was born Resource pooling technology further improves the reliability and availability of cloud computing. In both phases, optimizations are made based on traditional, CPU-centric computing architectures. Obviously, when Alibaba Cloud developed the cloud operating system Feitian, the optimization of the CPU was only a matter of time and specific paths.

A data from the International Data Corporation (IDC) shows that the global data volume has grown at a compound annual growth rate of nearly 50% in the past 10 years, and further predicts that the demand for computing power will double every four months. This means that with the failure of Moore’s Law, the performance of the CPU will reach the ceiling, and the processing demand after the increase in data volume may be unsustainable.

In addition, enterprises are increasingly using data-intensive computing application scenarios such as big data and AI. Huang Ruirui, chief architect of Alibaba Cloud’s basic products, also emphasized: “We found that there are many core applications on the cloud, which are actually data-intensive applications. After serving so many customers, we found that customers’ demands for us are to gradually change from business to business. Logic as the center, turned to data processing as the center.”

In fact, these are constantly increasing the requirements for low latency and high bandwidth provided by cloud computing. From Alibaba Cloud’s point of view, the CPU-centric computing architecture is difficult to adapt to, and faces great challenges: the CPU-centric architecture has low data throughput, large computing and network transmission delays, and cannot solve the problem of ultra-large scale. The increase of big data applications has led to an increase in the amount of data migration within the data center, and the CPU-centric architecture cannot provide high bandwidth.

It’s time for a change.

Alibaba Cloud’s idea is actually to make systematic innovations from the internal architecture of the data center, turning the CPU-centric architecture in the past into a cloud operating system + CIPU-centric architecture.

get?code=YzI3ZDQyYjQ0ZjE5NjYwZTBjYzJjMDV Figure | Alibaba Cloud CIPU Architecture Diagram

In this, the Feitianyun operating system runs on the CIPU, so as to obtain better management capabilities. Huang Ruirui further added: “We hope that millions of servers will be managed, orchestrated, and scheduled as a whole to become a supercomputer. This itself is also the goal that “Feitian + CIPU” wants to achieve.”

In other words, “Feitian + CIPU” is actually a transformation of the cloud computing data center architecture, while the CIPU is more like a big data center and a large cloud computing controller, supporting the cloud operating system to more efficiently manage cloud resources. Manage and accelerate.

02 Will the CIPU have more opportunities?

The CIPU is certainly not the only one. Around data center computing, there have been a series of innovations in chips and software, and even the evolution direction of different technologies.

For example, the DPU (Data Processing Unit) released by NVIDIA in 2020 is positioned as the “third main chip” after the CPU and GPU in the data center. Functionally, the DPU acts as an engine for computing offload, and the direct effect is to “reduce the burden” on the CPU. Later, in the direction of DPU, a number of start-up companies emerged. In accordance with the rapid growth of data throughput applications, Intel launched IPU (Infrastructure Processing Units) in 2021, integrating the experience of cooperation with cloud manufacturers, trying to improve the ability of data-intensive scenarios by using CPU.

In Alibaba Cloud’s view, the virtualization cloudification capability represented by IPU and the data relocation bandwidth capability represented by DPU are difficult to integrate. Only cloud vendors can truly achieve it and achieve scale. CIPU can break through the bottlenecks of the above two.

In the internal communication with Alibaba Cloud, I felt their strong self-confidence, because once you master the industry Know-How, you will have the opportunity to define the future. Jiang Jiangwei, head of Alibaba Cloud’s basic products, also emphasized, “The biggest difference in our CIPU is that we use the Feitianyun operating system defined at the top level to define the chip vertically based on business drivers.”

In the field of application design, a cloud native (CloudNative) concept has recently become popular. Under the cloud-native wave, the previous focus was on the infrastructure, focusing on PaaS, SaaS, and applications. But there should be another perspective, which is to look down, such as the upper-level requirements of data-intensive computing, which are also promoting the “nativeization” of the hardware side. Cloud native should include both software and hardware concepts.

In fact, in the combination of “CIPU + Feitian”, Alibaba Cloud first made the cloud operating system very mature. Through the definition and iteration of Feitian, a clear perception of customer needs has been achieved. After this, only to do CIPU. The CIPU built for the cloud can be deeply customized with the underlying infrastructure, which can more accurately and vertically solve the management problems of cloud operating systems. To a certain extent, in this process, the cloud nativeization of the CPU is actually completed.

What is the greatest meaning of originality? Perhaps it is to achieve the ultimate performance and experience. Jobs quoted Turing Award winner Alan Kay’s famous saying at the first iPhone launch: “People who are extremely serious about software should produce their own hardware.”

get?code=MTg5OGU3N2RiNzRlOWUyMmRjMjc3ZWY Figure | Jobs quoted Alan Kay’s famous words at the press conference

If you go back to Apple’s development, you will find that Alibaba Cloud’s “cloud operating system + CIPU” path selection is similar to Apple’s. Apple’s mobile phone first had an iOS operating system, and it actually used Samsung’s CPU at the beginning, and then used its own A-series CPU; Mac initially used Intel’s CPU, and finally became a self-developed CPU, such as MI and M2 chips. amazing. Through self-developed software and hardware integration, Apple has created a highly competitive user experience. Companies in two different fields are defining the future of their respective industries in similar ways.

Of course, there are superficial similarities and differences between the two.

As a terminal product, Apple needs to consider the interconnection and interoperability of complex ecosystems. The scenario used is human-computer interaction. This actually means that the product may not follow the principle of efficiency first because of the problem of path dependence, and even the improvement of efficiency must be accompanied by It is a huge obstacle, so it cannot achieve the “extreme” in the original sense.

For example, the original intention of the “QWERTY” keyboard was to solve the immature development of the early mechanical typewriter keyboard. When the typing speed is too fast, it is easy to cause the problem of “jamming”. As the inventor, American Christopher Shoals chose “anti-human design” to solve this problem by arranging commonly used letters in places that are not comfortable, thereby reducing the speed of typists. Although later with the advancement of processing technology, keyboards with more efficient alphabetical combinations appeared, but they did not become mainstream.

The “CIPU + Feitian Cloud Operating System” is different: first of all, it is oriented to the data center and is at the system level, not a stand-alone product like Apple’s products. Secondly, it is more to meet the interaction between machine and machine. Compared with the terminal, Alibaba Cloud is easier to implement native through CIPU + Feitian operation.

CIPU + Feitian operating system has greatly improved performance and experience in terms of scheduling efficiency and operating computing power, and the client does not need to make any code modifications to enjoy the benefits brought by this cloud computing power without feeling. accelerate.

In addition, from the perspective of Alibaba Cloud’s product layout, “CIPU + Feitian” can be deeply adapted with other self-developed products, such as Shenlong computing platform, Pangu storage platform and Luoshen network platform. In this way, better performance and even better price/performance ratio are achieved on the whole.

From another perspective, Alibaba Cloud has launched a cloud computing architecture system centered on the CIPU, which actually means that the company is trying to define a new generation of cloud computing standards. The future progress and possible changes here are quite worthy of attention.

Many people think that Ali’s e-commerce, payment and logistics are of great significance to China. I think this is actually only part of Ali’s influence on Chinese business.

Today, with digital and intelligent transformation sweeping across all walks of life, Alibaba Cloud’s new computing brings new business changes, which may be another important dimension of Alibaba’s impact on Chinese business.

Cloud computing seems to have entered a critical breakthrough period. Zhang Jianfeng, President of Alibaba Cloud Intelligence, said: “We see that cloud computing has entered a new stage of development. We are getting closer and closer to the next era, and we need to define a brand-new technology system. Therefore, our most important strategy this year is B2B, It’s Back to Basic, back to the essence of cloud computing.”

In my opinion, many times I want to truly define the future of the cloud computing industry. The first competition is the ability to think independently, especially the definition of future computing problems. And what Alibaba Cloud is doing deserves close attention.

This article is reproduced from:
This site is for inclusion only, and the copyright belongs to the original author.

Leave a Comment

Your email address will not be published.