Shulou(Shulou.com)11/24 Report--

On August 24, the Super converged Ethernet Network, jointly declared by Peking University and Huawei Technology Co., Ltd., won the solution Innovation Award at the National High performance Computing academic Conference (CCF HPC China 2023) held in Qingdao. CCF is the abbreviation of the Chinese computer Society. CCF HPC China is one of the three most influential supercomputing events in the world in the field of high-performance computing, with the title of "China supercomputing weather vane".

If the highest international award in the computer field is the Turing Award, then the highest domestic award is the award issued by CCF. According to people familiar with the matter, to be able to win the award awarded by CCF is undoubtedly a very high honor and strength of a computer discipline. The purpose of the CCF Innovation Award is to recognize outstanding achievements with important discoveries, inventions, innovations in computer science, technology or network engineering and international influence in related fields. Winning this award not only represents the ability of China's scientific and technological forces to creatively solve practical problems in computing, but also means that we can better share such innovations with the world in the future.

Jin Hai, vice president of CCF, presented the award (middle)

Fu Zhenxin, Computing Center of Peking University (left)

Zhang Bai, vice president of Huawei data center network field, receives the award (right)

So, what is the gold content of the joint innovation achieved by Peking University and Huawei? Let's go through it carefully.

Computer landing is faced with the problem of congestion.

In recent years, computing power, like water and electricity, has penetrated into all areas of production and life and become an important driving force for high-quality economic and social development. At present, the total scale of computing power in China ranks second in the world, maintaining an annual growth rate of about 30%. Intelligent computing power accounts for more than half of the new computing facilities, which has become a new engine of computing power growth. A series of remarkable growth points let people see the vitality that belongs to "Chinese computing power".

Since the end of last year, ChatGPT has attracted global attention to large models, which has ignited the demand for AI training in thousands of industries. The number of users of computing platforms in universities and scientific research institutes has soared, the amount of work of users has gradually exceeded the capacity of the platform, and the business throughput and complexity of network infrastructure have greatly increased. How to solve the congestion problem and let the computing power better serve the high-quality development has become a top priority.

Peking University chooses China as Super converged Ethernet solution

Peking University built a school-level public platform for high-performance computing as early as 2018, and has successively built and used five clusters, such as "Weiming No.1", "Weiming No.2" and "Weimingke No.1", with a total of 732 nodes and more than 31000 cores, calculating peak 3.65PFLOPS and equipped with storage 14PB.

The school-level public platform for high-performance computing improves the scientific research service capability of Peking University as a whole and reduces the cost of scientific research. As of May 12, 2023, the platform has provided high-performance computing services to more than 5000 teachers and students from 96 departments, including mathematics, mechanics, physics, chemistry, biology and geosciences, supporting 545 scientific research projects and more than 1400 scientific papers. And supported the release of the Goldenberg Prize in 2020.

However, with the increase in the number of users of the platform and computing tasks, the throughput and complexity of the network infrastructure have greatly increased, with an average annual node occupancy rate of more than 80%, and the node occupancy rate of "unknown student No.1" is even more than 95% for a long time. The job time and queue time are getting longer and longer.

The task of tackling key problems in scientific research should not be blocked in the face of the door. In order to improve the efficiency of high-performance computing and reduce the cost of scientific research, the school-level public platform of Peking University has organized a factory "assessment" to select the best high-performance computing network in the intelligent era. Through rigorous testing, the school-level public platform for high-performance computing at Peking University chose Huawei Super-converged Ethernet solution to build high-performance computing clusters based on intelligent lossless high-performance computing networks, releasing 100% of the computing power. reduce job running time and queuing time.

Why Huawei Super converged Ethernet solution stands out

In order to solve the problem of long queuing and working time of popular large-scale public computing platforms, various manufacturers have successively proposed IB (Infiniband), RoCEv1, RoCEv2 and other lossless network technologies. However, the traditional Ethernet deployed in the general computing area, the IB network deployed in the high-performance computing area and the FC network deployed in the storage area are three heterogeneous networks with different protocols and fragmented architectures. it brings some problems, such as difficult operation and maintenance, closed private network ecology, high cost, unable to achieve full life cycle management and so on.

Huawei Super converged Ethernet solution exclusively realizes Ethernet 0 packet loss. Compared with traditional Ethernet, Huawei can keep the server size unchanged and double the computing power; it can build a 10e-class super large-scale computing cluster; it can exclusively realize network computing and network-assisted computing. After being certified by the third-party authoritative testing organization Tolly, the task completion time (JCT) is reduced by 17% compared with IB.

The joint award by Peking University and Huawei is no accident, which proves the full recognition of Huawei's super-integrated Ethernet solutions in the technology circle of the supercomputing industry.

With the development of intelligent digital industry to intelligent world, data centers will evolve towards multi-cloud and heterogeneous computing power in the future. in order to build strong computing power, we need a high degree of coordination and deep integration of all kinds of resources. An open and independent super converged Ethernet network will coordinate multi-dimensions with all kinds of computing resources to comprehensively improve network performance and release higher computing power, thus contributing to the construction of a new computing center with "high computing power, high security, high efficiency and high reliability". Huawei's super-integrated Ethernet solution is expected to play an important role in national scientific computing, engineering and technological innovation, the use of sophisticated scientific research and computing power and construction, and add fuel to the high-quality development of "Chinese computing power".

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.