The relationship between the development of cloud computing technology and chip processor

CPU： a central processor , It represents people's demand for general computing power , Like smartphones , The computer , And the rise of the Internet and mobile Internet that they brought .

GPU： Graphics processor , It represents people's demand for high-performance parallel computing power . Such as ： Artificial intelligence , Big data applications

CIPU： Cloud infrastructure processors （ Alibaba cloud ）. Specialized processing and acceleration of cloud computing infrastructure

         Application of Cloud Computing , Such as ordering takeout by mobile phone , Watch videos and live broadcasts online , Mobile shopping is supported by cloud computing . Cloud computing and data center are the big blue ocean in the coming decades , It will be the next big cycle and trend of technology explosion after artificial intelligence . In recent years, many chip giant companies have been laying out in this field , Like Intel's IPU, NVIDIA DPU.

The development experience of cloud computing technology

         Cloud computing is to put computing on the cloud , Cloud refers to all kinds of complex hardware and software connected through the Internet .

         The emergence of clouds ： When the PC just came out , All kinds of software can be installed on it , But it can only be used by one person , After the network appears , Many people's computers can be connected through the network , You can transfer data to each other , They communicated with each other . But at this point , Processing data and communication depends on everyone's own computer . later CPU The processing power is getting stronger and stronger , There are more powerful computers than personal computers , It's the server . Network technology is also developing , The data transmission speed is getting faster and faster , In this way, people can use the server to handle most of the data and communication tasks , Finally, it is good to return the results through the network .

         As data and applications continue to increase , The number of servers needed is also increasing , In this way, data centers have emerged , Each data center is like a large computer room , There are millions of servers inside , They are connected by the Internet , This forms a small cloud , Data centers in many parts of the country and even the world , Connected by high-speed network , That's it A big cloud . We order takeout by mobile phone , Or watch live webcast , All need clouds to process data , And put the order results , The location of the takeout , Or the live video pictures are transmitted to us through the network .

         When demand changes , New technologies will emerge . In the past decade , The development of cloud computing has gone through two main stages .

Distributed technology

 

         The focus of the first stage , Distributed technology , It corresponds to centralized technology , Break up the computing tasks originally in a server or a computer room , And distributed to multiple servers and multiple data centers . The benefits of doing this , The most intuitive thing is to improve the stability of the system , It is equivalent to putting eggs in multiple baskets , Even if a server or data center suddenly hangs up , Nor will it affect the normal operation of the entire distributed system . Another benefit is improved performance , Because a large task can be decomposed into many small tasks that can be run at the same time , Let multiple computers process in parallel .

Resource pooling Technology

         Later it was found that , Distributed systems are not optimized enough , Because a node often has to deal with computing, network storage and other tasks , If you switch back and forth , Very trouble , It also has a great impact on performance , So the second stage of Cloud Computing Development , Is the emergence of resource pooling technology . This concept is more abstract , It's like a pond one by one , The water in this pond is specially used for watering crops , The water in another pond is used to raise fish . In Cloud Computing , Resource pooling is based on distributed computing , Built a lot to do calculations 、 Storage pond , Also called a resource pool . Each resource pool is actually a bunch of servers , But these servers are specially used for computing or storage . So through this architecture of computing and storage separation , It can efficiently and uniformly schedule the resources of the whole cloud , There is no need for another server to server control , This further improves the reliability and flexibility of cloud computing .

         The two stages have a common feature , That is, they are all defined by software , Put computing power , Also known as computing power, it becomes a public computing resource . A lot of people think that , Like electricity, computing power will become an indispensable resource in the future . For example, our country has been building a system of counting from the east to the West , And the previous South-to-North Water Diversion Project , West to East power transmission , The computing power has been raised to the strategic level of resources .

Changes in processor requirements

         Now the demand changes , We order each time , Every time you watch a video, you will generate a vast amount of data , This kind of data intensive computing is more and more , It has caused great pressure on the infrastructure of cloud computing . In the final analysis , Because the current cloud infrastructure is based on CPU Core , And modern CPU It is difficult to meet this demand .

         Is not to say that CPU That won't be possible , Is too weak ,CPU Still the most important chip of all , Not one of them. , So traditional computers 、 Including servers and data centers , Their hardware and software architectures are based on CPU Designed for the center , But in the face of the new demands arising from the big data and cloud computing changes mentioned above , This kind of CPU The center - centric architecture is a little weak .

First of all , Not good at parallel computing and network control

        CPU Processing of instruction sets , Not good at parallel computing and network control , This is why and AI Relevant calculations should be made with GPU, The reason why network card chips are needed for network transmission related applications .

second , It is difficult to provide high bandwidth .

         Big data is bound to be accompanied by a large amount of data handling and migration , This brings the demand for high bandwidth , This is also CPU Unsatisfied .

Third , It's hard to balance complexity 、 Time delay 、 performance .

         Now the scale of cloud computing is very large , Hundreds of thousands of servers , This increases the complexity of the whole system management .CPU As a general chip , It's hard to balance complexity 、 Low latency and high performance .

Market outlook under new demand

         Many chip companies have also encountered such problems , And give their solutions . For example, Intel put forward IPU, That is, infrastructure processors . And NVIDIA also quit DPU That is, the data processor . The significance of these two chips , To better meet the new needs of cloud computing , Improve the throughput of data processing 、 Reduce delay , And directly support virtualization and cloud from the hardware , These are all CPU perhaps GPU Impossible .

        IPU and DPU The essence is similar , All right CPU Complement and extend the core cloud computing architecture . They can CPU What I'm not good at next , Transfer to yourself to complete more efficiently , This reduces CPU The pressure of the . But these chips have not changed the fundamental problems mentioned above , As long as you still CPU At the core , So those complexities 、 bandwidth 、 The problem of inefficiency remains . We can see that , These solutions all look at this problem from the perspective of chip and hardware . But the revolution in Cloud Computing , In fact, it is driven by software definition .

         No software , Chips are just expensive sand . Perhaps the fundamental solution to these problems , It starts with software , Rethink the architecture of the future cloud data center .

source ：

Alibaba cloud has developed a new processor , And grab CPU My job ？_ Bili, Bili _bilibili