Peter the Great St. Petersburg Polytechnic State University
SPbPU's supercomputer center is focused on solving inter-disciplinary natural science research tasks and design of complex technical systems for hi-tech segments of local industries and science.
After scheduled upgrade of ‘Polytechnic - RSC Tornado’ supercomputer in 2020 its peak performance has increased by 29% and the system got 28th place in the new edition of global IO500 rating of the best performing HPC-level storage systems (as it was announced at SC20, the world’s largest supercomputer exhibition).
Hyperconvergent architecture
Hyperconvergent systems combine computing and storage resources on each node. Each node is a part of both computing and distributed storage subsystems (compute/store workloads). This enables linear scalability of system resources with the increase of the number of nodes.
Unlike classic HPC approach where computing and storage resources are scaled separately, in hyperconvergent system you can increase both computing power and distributed storage capacity/speed as you increase the number of nodes.
Supercomputer software based on RSC BasIS supports configuration and management of such systems allocating computing and storage resources for each user workload.
Distributed On-Demand Storage
Hyper-converged approach resulted in unique high-speed storage system RSC On-Demand Storage for Polytechnic RSC Tornado supercomputer with leading characteristics - total parallel file system speed is up to 194.9 GB/s.
The system is ranked 28th in the current edition of IO500 list (November 2020), the new industrial rating for HPC-class storage systems.
RSC Tornado servers as compute resource and distributed storage system
RSC Tornado servers TDN511 with direct liquid cooling are well balanced for various computing tasks and as a part of RSC Storage-on-Demand system supporting different file systems.
Distributed storage is a centrally managed system with multiple data layers (for “hot”, “very hot”, "warm" and "cold" data).
Very hot data storage system is based on four RSC Tornado TDN511 blades. Each server has 2 high-speed low-latency Intel® Optane™ SSD DC P4801X 375GB M.2 Series drives with Intel® Memory Drive Technology (IMDT) providing 2.8 TB for “very hot data”.
Hot data storage system includes dynamic RSC Storage on-Demand with parallel Lustre file system based on 62 RSC Tornado TDN511 blades. Low latency Intel® Optane™ SSD DC P4801X 375GB M.2 Series drives are used for fast access to file system metadata and Intel® SSD DC P4511 4 TB (NVMe, M.2) drives are used to store “hot data”.
Warm data storage system is a static hard disk storage with parallel Lustre file system providing 1 PB for warm data.
Cold data storage system includes 2.4 PB NAS storage and tape storage system.
High-speed Mellanox InfiniBand FDR network
Mellanox InfiniBand FDR technology with up to 56 Gbps unblocked switching speed per port is used for high-speed data transfer between computing nodes in Polytechnic RSC Tornado supercomputer. Adapter with 2 InfiniBand FDR ports is installed in each blade to increase network access speed to 100 Gbps. This provides required performance for demanding user workloads and distributed storage system of a supercomputer.
Energy efficient solution
Liquid cooling of Polytechnic RSC Tornado minimizes cooling costs of the most powerful servers. Average PUE factor (power usage efficiency) is less than 1.2. It means that the cooling system consumes less than 20% of total consumed power, which is an outstanding result for HPC industry.
