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

Since the beginning of the 21 century, human scientific exploration has gradually gone deep into the new environment with high complexity and high uncertainty. A large number of large and complex scientific problems need to be simulated, simulated and predicted by supercomputers. From the traditional chemical and chemical industry, bio-pharmaceutical, high-energy physics, engineering, social governance, to the emerging big data, artificial intelligence, network information security and so on, they have made more new progress and achievements than ever before due to the blessing of strong computing power.

Peking University Computing Center, founded in 1936, is one of the earliest computing centers in Chinese universities, and a forerunner in the establishment of a high-performance computing school-level platform, providing a strong support for scientific research. In 2023, the Computing Center of Peking University celebrates its 60th anniversary.

Scientific research and innovation ushered in the era of computing power

In 1963, when Peking University had its first computer, it took the lead in setting up a computing center in colleges and universities across the country and set up a professional and technical team to serve all teachers and students. Half a century later, supported by the technical team of the Computing Center, Peking University established and launched the school-level public platform for high-performance computing of Peking University. Subsequently, "Weiming No.1", "Weiming Teaching No.1" and "unknown students No.1" and other clusters have been put into operation one after another.

The main function of the school-level public platform for high-performance computing of Peking University is to provide high-performance scientific and engineering computing services for the whole university to meet the needs of large-scale data processing and large-scale scientific computing in various disciplines. The main task of the platform is to build a scientific and reasonable management system, an architecture for sustainable development, a professional and stable technical team, an efficient and advanced computing environment, rich and comprehensive software resources, infrastructure with perfect energy saving and environmental protection functions, reasonable and convenient charging mechanism, cooperative and shared operation and maintenance system, to realize the resource integration and efficient use of school-level platform and college-department secondary platform. Improve the scientific research service ability of the school as a whole and reduce the cost of scientific research.

At present, the total number of computing cores of the public platform is 14064, the peak computing capacity is 1680Tflops, and the storage capacity is up to 11084TB, which has become an important scientific research supporting environment of Peking University. The platform provides a high-performance computing environment for mathematics, mechanics, physics, chemistry, biology, geoscience, computer science and other disciplines. As of May 12, 2023, the high-performance platform of Peking University has a total of 5070 teachers and students, distributed in 96 faculties and departments of the university. at the same time, it is known to support more than 545 scientific research projects, with a total amount of 3.136 billion yuan, and more than 1400 scientific papers, and supported the release of the Goldenberg Prize in 2020. The award-winning project, which raises the molecular dynamics simulation limit to a staggering number of 100 million atoms through machine learning, is considered to be the most exciting development in the field of computing science today.

Speed up is imminent, Huawei intelligent lossless high-performance computing network helps release the ultimate computing power

With the increase of the number of users of the platform, the amount of work of users gradually exceeds the capacity of the platform, and the traffic throughput and complexity of network infrastructure are greatly increased. It is urgent to transform the system and network of Peking University's high-performance computing school-level public platform.

Huawei Super converged data Center Network CloudFabric 3.0 solution builds high-performance computing clusters based on intelligent lossless high-performance computing networks, frees up computing power, and helps Peking University supercomputing platform shorten job running time and queuing time.

It is understood that Huawei Intelligent Lossless High performance Computing Network focuses on the evaluation results of TCP / IP, IB and RoCEv2 in the application scenarios of high performance computing benchmark tool Linpack, Earth system model CESM and molecular dynamics software VASP, respectively.

The experimental results show that in VASP test, Huawei intelligent lossless high performance computing network 100GE RoCEv2 is better than IB;. In Linpack and CESM tests, Huawei intelligent lossless high performance computing network 100GE RoCEv2 is basically consistent with IB performance, which also confirms that Huawei intelligent lossless high performance computing network can replace IB in real application scenarios.

Huawei intelligent lossless high performance computing network 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 provide CloudEngine 16800, the industry's highest density 768*400GE switch, and can build a 10e class super large computing cluster. Exclusive implementation of network computing integration, network-assisted computing, certified by the third-party authoritative testing organization Tolly, task completion time (JCT) is 17% shorter than IB.

The high-performance computing school-level platform of Peking University has the leading supercomputing cluster in China, and the Linpack efficiency of the whole system ranks in the forefront all the year round, which requires very high network performance, reliability and advanced nature. The test results deepen the full recognition of Huawei's super-converged data center network with lossless Ethernet in the supercomputing industry.

In recent years, Huawei super-converged Ethernet solution has been applied in many high-performance computing platform projects at home and abroad, and has been praised for its superior performance. At the same time, Huawei RoCE lossless Ethernet network is also making continuous progress in terms of delay, throughput, reliability, codec and modulation technology, so that the overall performance of lossless Ethernet network continues to improve. In the future, it is expected to be used in autopilot, intelligent manufacturing, biotechnology, and other industries and fields, contributing to national scientific computing, engineering and technological innovation, and advanced scientific research.

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.