The Institute of Computational Mathematics and Scientific/Engineering Computing of Chinese Academy of Sciences was founded in 1995 and it originated from the former Computing Center of Chinese Academy of Sciences founded by Professor Kang Feng. The Institute aims to address major application issues in science and engineering, and focuses on the theoretical and technical innovations in fundamental and critical computational methods. It is a multidisciplinary research center for basic computing method research.
The Institute examines finite element and boundary element method, multi-scale computation, linear and nonlinear optimization, computational hydrodynamics and parallel computation. With the development of computer technology, the Institute needs to engage in HPC programs and software R&D that reflects the latest research findings in the international computation sector, and nurture a full team of high-caliber talents who could adapt to current scientific and engineering computation needs. Therefore, it needed to establish a HPC cluster urgently.
Analysis of application features
CAE application features can be divided into invisible finite element analysis (IFEA), evident finite element analysis (EFEA) and computational fluid dynamics (CFD). The basic features of IFEA include high memory occupation, big disk IO and high bandwidth requirement of memory. Therefore, it requires a system with high memory capacity, high access bandwidth and fast disk IO speed. Comparatively, EFEA and CFD software have a lower requirement on memory and disk IO but higher requirement on communication delay. In general, a high-speed network is selected. Most manufacturers and scientific research institutes have diverse application needs for CAE software and need to have evident and invisible finite element analysis and fluid dynamics computation simultaneously. Therefore, the systems solution needs to consider two different application modes.
Considering the requirements of the large-scale HPC cluster of the Institute of Computational Mathematics and Scientific/Engineering Computing of the CAS, Inspur specially launched Tiansuo TS10000 HPC cluster solution for this project. Inspur’s system solution entails high single-point configuration, 20Gbps InfiniBand computational storage network, high-performance core 1Gb Ethernet monitoring and dispatching network, high-efficiency power source (a power conversion ratio of 93%) and uniform remote cluster management.
The cluster server comprises 282 dual-way blade computing nodes, 1 4-way fat node, 1 dual-way high-level management node, 2 dual-way login nodes and 2 dual-way IO nodes. The system has a peak computing capacity of 24.1 TFLOPS;
The cluster storage features one Inspur NF5280+SAS storage with massive storage spaces. It is furnished with 2*146G SAS + 28*450G SAS 15000rpm hard disks with a net capacity of 12.6T. The available capacity after Raid5 is 10.8T (including two disks for hot standby but not for Raid);
The cluster computing network comprises Mellanox DDRIB-288 core InfiniBand switch which provides 288 ports, redundant power supply and fan design.
The cluster management network comprises H3C 7510E core 1G switch which provides 336 ports (can be expanded to include 480 1G ports), redundant power supply and fan design.
Intel C, C++, Fortran, Java encoder as optimized for the Intel platform, Intel MKL mathematic function library, Intel VTune optimization tool, Intel Trace analyzer, Trace Collector analytical collector and MPI, mpich, OpenMP, ACML, BLAS, LAPACK, Scalapack and FFT are provided as part of the cluster’s parallel application environment.
Highly efficient computing method and high-speed computer are equally important. As an important means for understanding and reforming the world, computation has become one of the three pillars of modern scientific research — together with theoretical analysis and scientific experiment. Tiansuo TS10000’s successful application in the Institute plays an important role in promoting China’s research on large-scale scientific computation and new scientific computation modes. During project implementation, Inspur’s project team communicated with the customer closely and formulated detailed action plans. The project won high compliments from the customer as a result. In the words to the Head of the Institute, “the careful and responsible attitude of Inspur’s engineers is really laudable. This systems solution satisfies our current application requirements. The whole cluster features high scalability and the system is open and easy to operate. Inspur TS1000 high-performance server delivers high performances and it is easy and convenient for operation and management.”