At the 200215th Future Network Development Conference, experts and industry leaders from industry, academia, research institutions and other fields discussed the key issues and changes of new network technologies around hot topics such as network operating system, 6G communication, network security and industrial Internet. Xu Wenwei, director of Huawei and president of Strategic Research Institute, delivered a keynote speech on future network research and breakthrough. In particular, he called on: "We should strengthen the cooperation between Industry-University-Research, promote future network research by means of mathematics and system engineering, and drive theoretical research and technological innovation by scene demand. Together with our partners, we have carried out the research on deterministic WAN and made good progress. The next step will accelerate the transformation of results and inject new kinetic energy into the development of advanced industrial networks. "
Embedded Niubizi Network Security Network Technology Industrial Network
How to better realize industrial interconnection in the future?
Mosaic ox script
The following is the full text of the speech
1. Understand the future challenges and directions from the three elements of the world.
Material, energy and information are the three elements of the world, and they are also the starting point for us to grasp the future challenges and directions. In the next decade, the number of connections will reach 1000 billion, the broadband speed will reach 10 Gbps per person, the computing power will increase by 100 times, the storage capacity will increase by 100 times, and the use of renewable energy will exceed 50%. Around the generation, transmission, processing and use of information and energy, technology needs to evolve constantly.
2. Smart machine connection in industrial interconnection: 100 billion nodes, trillion market.
In recent ten years, the vigorous development of consumer Internet has greatly enriched people's communication and life. At present, the Internet penetration rate calculated by population has reached 70%. Looking forward to the future, with the improvement of 5G2B capability and the large-scale deployment of IPv6 protocol, the number of smart phone connections will usher in explosive growth. According to ARM's forecast report, it is estimated that the cumulative number of machine connections in the world will reach 654.38+000 billion in 2035.
Consumer Internet mainly provides people with connections, but people's sensory response speed is limited, so people are more tolerant of the experience fluctuation caused by network quality fluctuation. For example, consumers can accept the switching speed of TV channels within 200ms, and it is not easy to detect the interruption of voice communication within 200 ms. The machine-based industrial Internet requires deterministic bearer with low delay and low jitter. AGV automatic guided vehicle requires industrial network delay less than 50ms and jitter less than 10ms, and some industrial control scenarios in steel industry require industrial network delay less than 4ms and jitter less than 50us.
3. It is a trend that few people are on duty and nobody is on duty, and the production mode of remote centralized control needs deterministic WAN.
We see that there are fewer and fewer production systems in enterprises, and the scene is unmanned. It is an inevitable trend to reduce costs and increase efficiency, and safe production. Industrial control system is accelerating the development to remote centralized control mode, which allows operators to complete production tasks in a safer and more comfortable centralized control room, and also enables large enterprises to realize the scheduling and optimization of production factors between headquarters and multiple bases in a wider range.
Therefore, industrial control system needs wide area, and industrial network needs to ensure certainty and large network scalability at the same time. The core technical challenge is to reduce the storage and computational complexity of deterministic network scheduling algorithm, so that it can be deployed in large-scale networks.
4. Facing industrial interconnection, improve the certainty, security and flexibility of IP network.
Today, the main body of network support is tens of billions of consumer interconnection. In 2030, the main body of network support is trillion-level industrial interconnection, and network protocols are facing triple tests.
The first is certainty. We need to ensure the ability of deterministic delay, and through the "new theory and protocol of network calculus", we can change the current best-effort network delay into deterministic delay that can be calculated in advance.
The second is safety. In the context of the Internet of Everything, the security defense system is facing severe challenges. A large number of external devices, such as drones, cameras, edge computing, sensors, etc. New insecurity factors are introduced, so it is necessary to build an end-to-end endogenous security framework and protocol.
The third is flexibility. The needs of various industries are diverse, some need longer IP addresses, and some need shorter IP addresses. It is necessary to expand the fixed-length IP address into a new IP protocol that can flexibly define semantics and syntax.
The modeling and optimization of deterministic WAN involves the queuing theory of network layer, which is complicated. We hope to provide some help with the help of mathematical tools and system engineering methods.
5. Network deterministic modeling and quantitative analysis based on network calculus theory.
Network calculus theory is an important mathematical tool for network modeling and time delay analysis. It describes the envelope curve of traffic arrival and node service, and calculates the upper bound of end-to-end deterministic delay based on minimum algebra. Professor Cruz of UCLA University invented network calculus in his paper 199 1, which was further sorted out and perfected by Swiss scientist Boudec.
Minimum addition algebra is a mathematical tool used to model discrete event systems, such as digital circuits, communication networks and manufacturing industries.
Based on the network calculus theory, due to the reuse of IP networks, streams will collide. Therefore, after the flow passes through the network node, the burstiness will increase and the delay and jitter will increase. However, in the traditional IP aggregation scheduling, the collision between flows or even cyclic blocking will lead to burst and delay upper limit. With the snowball-like open growth of hops, even if the network is lightly loaded, there will be a big delay.
6. Further optimization through system engineering method to solve the uncertainty of large-scale WAN.
In order to solve the problem of delay uncertainty in IP network, the industry has done a lot of research for decades and put forward many technologies with certainty guarantee. However, these technologies basically rely on maintaining the flow state in network nodes, or the scheduling complexity is very high and the scalability is not good, so they cannot be applied to large-scale networks.
In order to solve the problem of certainty and scalability at the same time, we adopt the methods of system engineering and periodic aggregation scheduling to maintain the state at the edge nodes and carry it in the message to avoid the state explosion of the core nodes. The core node only needs to maintain four aggregate circular queues, and the scheduling complexity is O( 1), which realizes the scalability of calculation and storage.
7. Optimization effect: E2E delay has a linear relationship with the number of hops, while delay jitter has nothing to do with the number of hops.
Theoretically, compared with traditional IP, the delay increases superlinearly with the increase of hop count. In our deterministic WAN technology (DIP), the upper limit of end-to-end delay is linear with the number of hops, and the end-to-end jitter is constant regardless of the number of hops. In engineering, we have the ability to reach the upper limit of single-hop delay of ten microseconds and the upper limit of end-to-end jitter of 10us.
8. Based on CENI test network environment, the deterministic WAN technology test is completed.
Based on the network calculus theory, the industry has made some breakthroughs in deterministic WAN technology.
In June, 2020, Huawei cooperated with Purple Mountain Lab to conduct deterministic WAN technology test on CENI network. In a 3000 km 13 hop network, the delay jitter is controlled within 100μs. However, the delay jitter of traditional IP traffic without deterministic WAN technology reaches 2.8 ms It should be noted that the delay jitter is controlled within100μ s, and the delay is related to the distance and cannot be reduced.
When the background interference traffic continues to increase, the delay jitter of deterministic WAN remains unchanged, while the delay jitter of traditional IP traffic increases with the increase of background traffic.
9. Based on the deterministic WAN technology, the world's first wide-area cloud PLC scenario technology test was completed.
Over the past year, Industry-University-Research has further carried out cooperative innovation, applied deterministic WAN technology to industrial network scenes, cooperated with Zijinshan Laboratory, Shanghai Jiaotong University and Baoxin Software to carry out wide-area cloud PLC cooperation, and completed the world's first wide-area cloud PLC scene technology experiment.
In the experiment, Yunhua PLC was deployed in Shanghai and controlled to be deployed in Nanjing 600 kilometers away through CENI deterministic WAN connection. Under the impact of large background traffic, the wide-area cloudization PLC system runs normally and stably. In the experiment, the network delay is less than 4ms and the delay jitter is less than 20us.
It is worth mentioning that this experiment adopts the integrated development environment and running environment of Kunpeng CPU, Euler operating system and PLC based on IEC 6 1499 standard provided by Shanghai Jiaotong University. Based on general computing and standard IP protocol, the industrial control edge computing node is constructed, and the prototype of full-stack positioning system is realized. I believe that in the process of upgrading to the next generation industrial control architecture, domestic universities and research institutions will play a greater role in thoroughly solving the problem of seven countries and eight systems in industrial control system.
10. Deterministic WAN enriches the IP technology capability set in a limited domain and accelerates the IP of industrial networks.
Deterministic WAN is the result of network calculus and system engineering optimization in a limited domain, which is oriented to specific industrial scenarios and enriches the IP technology capability set, thus serving more industrial network scenarios. New technical capabilities such as deterministic WAN, combined with IPv6 and 5G2B, will accelerate the IP of industrial networks and further expand the service scope of consumer+industrial Internet. We can refer to the practice of 5G and provide different capability sets for different scenarios, such as eMBB and uRLLC.
The demands of consumer Internet and industrial Internet are different. It can continuously enrich the IP technology capability set, make the consumer+industrial Internet interconnected based on the common capability set, and at the same time serve more scenarios based on the enhanced capability set in a limited domain, and continuously expand the service scope of the Internet.
1 1. Explore the certainty of end-to-end systems: from deterministic networks to deterministic systems.
Based on the research of deterministic WAN, now we are also exploring a complete end-to-end deterministic system. For example, technologies such as deterministic 5G air interface, deterministic WiFi, deterministic industrial field network with extremely low delay, deterministic host protocol stack, etc. need to apply mathematical theories such as random network calculus and martingale theory, and also use system engineering optimization methods to meet the service delay in a deterministic or probabilistic way with the least network resources. Another example is to explore E2E's deterministic system for key application scenarios, not only deterministic networks, but also deterministic computing, operating systems, chips and cloud services.
12. Mathematics and system engineering methods can play more roles.
Mathematical tools and system engineering methods can play more roles in the future network development.
For example, network-level MBB experience optimization can achieve the best network energy efficiency and user experience with the help of "Shannon capacity model at network level". In this way, Huawei assisted the sunrise in Switzerland. Under the disadvantage of spectrum and number of sites, the 5G network ranks first, and its experience is more than 1.5 times that of the second place.
For another example, in the energy consumption optimization of base station cooling system, through system scheduling optimization, theoretical analysis predicts that air conditioning energy consumption can be reduced by 30% and base station operating cost can be reduced by 10%. Next, we need to verify the actual effect in the actual environment.
In addition, in the aspect of optimizing the computing efficiency ratio of data center, the computing efficiency ratio can be significantly optimized by optimizing the data center network, realizing zero network packet loss and reducing CPU empty consumption caused by data retransmission. This will also involve mathematical methods such as graph theory and system engineering optimization method.
13. Develop forward-looking cooperation oriented to the future and accelerate the process of using Industry-University-Research.
To meet the needs of human development and solve the problems it faces, we need to pool the wisdom and innovative ability of all mankind, and we must overcome the challenges with an open, inclusive and coordinated innovation mechanism. Industry must work closely with universities and scientific research institutions to lead scientific research with the challenges of industry and world-class problems.
Facing the uncertain innovation in the future, we will carry out forward-looking cooperation from three levels to accelerate the application process of Industry-University-Research.
On the first level, the cooperation oriented to basic scientific research, through the strategic cooperation of talent funding, gift funding and other measures, supports global outstanding talents, focuses on the research direction of ICT basic technology, and promotes the breakthrough of basic theory.
The second level is the cooperation of basic technology research. Through joint laboratories, innovative laboratories, school-level framework agreements and other measures, with a cycle of 5 to 10 and an annual scale of one million dollars as the baseline, we will unite world-class research talents to build a world-class advanced technology laboratory, carry out cutting-edge research in the ICT field, and create and invent major new technologies.
The third level is the technical innovation cooperation of the project. We have launched projects such as HIRP Opening, Fellow Trial and Error, and Advisory Committee, which combine the advantages of scientific research in colleges and universities with enterprise problem scenarios, explore multiple paths, deepen cooperation and break through key technologies.
We hope to carry out technical research and innovation based on theoretical research breakthroughs in academic circles to create products and solutions that better meet the needs of users; The ever-developing needs of users and future scenarios will also urge us to pay more attention to theoretical breakthroughs and cooperate in research and development of new products, thus forming a good cycle of Industry-University-Research, promoting industry development and social progress, truly realizing our vision, bringing the digital world to everyone, every family and every organization, and building an interconnected world.