Peng Cheng Laboratory

Peng Cheng Cloud Brain II Captured Top Spot on IO500 List

Date: 2020-11-19 Source:PCL

The latest IO500 List released at the International Conference for High Performance Computing, Networking, Storage, and Analysis (SC20) on November 19 marked a significant milestone for the Chinese supercomputing systems. For the first time, a supercomputing system developed in China grasped the top post of the list. The “Peng Cheng Cloud Brain II”, a new system jointly developed by Peng Cheng Laboratory (PCL) and its partner institutions and enterprises across China, ranked the first on both the IO500 Full List and the 10 Node Challenge List, scoring 7043.99 and 1129.75, respectively. The powerful data throughput of the Peng Cheng Cloud Brain II as shown laid a solid foundation for a wide range of applications in artificial intelligence.

IO500-1.png

IO500 Full List


The performance of the storage systems is drawing increasing attention in terms of the design of large-scale supercomputing systems. IO500, as one of the world’s most authoritative rankings for storage performance in high-performance computing, has been widely recognized by commercial enterprises and international research institutions including top universities and national laboratories since the list’s inception. Since November 2017, the IO500 lists are released at ISC held annually in Frankfurt, Germany, and SC held annually in the US. 


Extreme testing and tuning to benefit the performance of system software and hardware were jointly conducted by PCL, Tsinghua University and Huawei Technologies Inc. in accordance with IO500 benchmarks including both data bandwidth BW (GiB/S) and metadata performance MD (KIOP/S). Test results demonstrated that Peng Cheng Cloud Brain maintains good scalability and stability at a variety of scales. The storage system of Peng Cheng Cloud Brain II is based on the supercomputer file system cache MadFS developed by the Department of Computer Science at the Tsinghua University. Innovative schemes including a highly scalable concurrent access based on the RUST programming language, large-granularity data caching/bypass access, data access/platform streaming, and zero-copy RPC extreme fast processing, have been developed accordingly, taking into account the hardware environment of Peng Cheng Cloud Brain II.

IO500-2.png

10 Node Challenge List

The two IO500 championships of Peng Cheng Cloud Brain II are attributed to the practical-use oriented design principle. Such a system not only quadrupled the previous champion’s full system score, but also outperformed its predecessor by nearly 50% on the 10 Node Challenge List which highlights the best storage performance for small data throughput, becoming the first system to exceed the score of 1,000 on the List. This means that Peng Cheng Cloud Brain II can provide world-leading data throughput and cutting-edge computing power for all scales of artificial intelligence applications. Peng Cheng Cloud Brain II, developed and placed into trial operation in October 2020, is now capable of providing more than 1000P OPS AI computing power for a wide range of applications from basic research to industrial empowerment.


AIPerf500-3.jpg

Peng Cheng Cloud Brain II


Submit Message

Scan to Follow Up

Peng Cheng Laboratory | Disclaimer | 粤ICP备18066427号