The HPC data storage system at the Poznan Supercomputing and Networking Center has again been highly ranked in the prestigious IO500 competition held during the ISC High Performance 2024 conference. The system took 8th place in the ’10 Node Production List’ category, in which it was listed for the first time, and 14th place in the ‘Production List’ category, in which the PSNC system has been present since 2023.
The explosion of big data, data science and AI applications means that IT systems, and in particular HPC computing and cloud systems, the construction and maintenance of which are at the heart of PSNC’s activities, require increasingly rapid access to data, use larger data sets and more files. Efficient data access is therefore key to the efficiency of computing, data analysis and progress in data-driven research, education and business.
At Europe’s largest HPC conference, ISC High Performance (13-16 May 2024), the TOP500 and Green500 lists – a list of the world’s fastest and most energy-efficient HPC computers – and the IO500 list – which compares the fastest data storage systems for HPC systems and data centres – were announced.
The PSNC data storage system for HPC was again highly ranked in the IO500 competition, coming 8th in the ’10 Node Production List’ category, in which it was listed for the first time, and 14th in the ‘Production List’ category, in which it has been present since November 2023.
The positions of the PSNC systems place us among the fastest producing academic and commercial solutions, alongside high-profile European partners such as LRZ, CINECA and UCL.
In the ’10 Node Production List’, the PSNC system is the highest ranked system based on the open source Lustre software, competing with closed (appliance) solutions from major storage system vendors, including DDN Exascaler, Intel DAOS, IBM Spectrum Scale and WekaIO.
The position of the PSNC system is even more successful because the system architecture and configuration were completely defined and optimised by the data storage system architects at PSNC. In developing the production and benchmark configuration of the system, we used the knowledge and experience of data processing system scalability and performance gained from many years of system maintenance and data management services provided by the PSNC Data Processing Division.
In addition, the ’10 Node Production List’ category of the IO500 ranking compiles the performance of systems for a specified maximum number of client nodes, allowing reliable comparison of the performance of systems of similar size. Such a ranking specifically promotes and rewards systems that have been appropriately designed and optimised within a pre-established scale of the competition rules. Again, the success of the PSNC solution should be highlighted, as it has been possible to build a system with performance comparable to the fastest, highly optimised vendor systems in an open system architecture.
Technical information
The system entered in the IO500 competition is based on the open Lustre software version 2.15.3 and the xFusion/Huawei server and array platform together with the InfiniBand network switches (Mellanox/nVidia) purchased by the PSNC in 2023.
In the current edition of the IO500 competition, we achieved a very high 8th place in the ’10 Node Production List’ category, thanks to the use of a next-generation InfiniBand network with a capacity of 200GBit/s.
Another key aspect of the PSNC’s system configuration is the extensive use of flash memory for high performance operations on metadata and the data itself. For the Lustre system’s metadata, we use full flash arrays, which ensure efficient execution of file operations, directory listing, access to file attributes, etc., while for the system’s data we use flash-accelerated disk arrays, which speed up access to multiple files and data blocks.
Summary
The HPC data storage system was procured and implemented by PSNC as part of the National Data Storage (NDS), PRACE-LAB and PRACE-LAB2 projects. The NDS project provides almost 700 PBytes of data storage on flash and disk storage, file and object-based systems and tape systems across 9 sites. The PRACE-LAB and PRACE-LAB2 (Cooperation in Advanced Computing in Europe) projects have designed and implemented HPC computing systems based on CPU and GPU platforms at 5 sites in Poland, together with specialised high performance data storage systems.
PSNC was the coordinator of these projects, which were carried out in cooperation with 8 partners, including HPCCs and MAN centres in Poland: National Centre for Nuclear Research (NCBJ), Centre of Informatics Tricity Academic Supercomputer and Network (CI TASK), Wroclaw University of Technology – Wroclaw Centre for Networking and Supercomputing (WCSS), ACK CYFRONET AGH, Bialystok University of Technology, Czestochowa University of Technology, Kielce University of Technology and Lodz University of Technology.
More information:
– IO 500 10 Node Production ISC24 List,
– IO 500 Production ISC24 List.
