Catalan High-Performance Computing center chose Bull to improve its services aimed at the scientific community and accelerate research
14 research groups have already tested the new Bull supercomputer in areas as diverse as chemical engineering, materials science, astrophysics and pharmacology.
Since mid-May 2009, users at CESCA (the Catalan intensive computing center) have had access to a new Bull supercomputer for calculation and computer simulation. The new system, code-name Prades that includes 240 processing cores, has significantly enhanced CESCA’s computing capacity, which now delivers up to 4.13 Teraflops of peak power. Thanks to this new supercomputer, users have an ultra-fast cluster system at their disposal, with shared memory, and an InfiniBand network.
Once the installation and benchmarking were completed, even before the system went into full production, users were able to see for themselves how the actual system performed. In total, the 14 research groups used over 130,000 hours of computing time in areas including chemical engineering, materials science, astrophysics and pharmacology. The scientists involved come from various research centers and universities, including CSIC (Consejo Superior de Investigaciones Científicas, the Higher Education Scientific Research Council), the Fundación Observatorio Esteve Duran, IMDEA (the Madrid Institute of Advanced Studies), the University of Barcelona, the Polytechnic University of Catalonia, the Autonomous University of Barcelona and the University of Gerona.
Key features of the Bull supercomputer
The new Bull system is a cluster of 14 Bull servers, featuring 28 nodes and 56 Intel® Xeon® 3.0 GHz quad-core processors (making a total of 224 cores). The system also includes a dual-node server dedicated to system management and Input/Output control; The cluster is capable of 2,688 million floating point operations a second, representing 2.7 Teraflops of power.
The ultra-compact Bull ‘twin’ server features Intel® Xeon® quad-core 5400 series technology (Harpertown). It delivers an optimized price/performance ratio and has been specially designed to be integrated as a compute node in intensive computing clusters.
A cluster featuring an ultra-fast InfiniBand network
Prades supercomputer features three networks: two Gigabit Ethernet (GbE) networks – one for system management and the other for service – with the third being an InfiniBand network dedicated to processing, enabling optimal performance for parallel applications. The GbE networks include a 48-port gigabit switch which includes one 10 GB port. Two InfiniBand switches have been installed to support the InfiniBand network: one for internal management tasks and the other for external tasks. In addition, each node features a dual-port InfiniBand board.
The Bull Advanced Server for Xeon software designed specifically for managing HPC cluster environments enables thousands of nodes to be administered. The software incorporates RedHat Enterprise Linux 5 and Bull System Manager HPC Edition, and enables the cluster to be effectively administered and controlled using a powerful tool, via a Web interface which simplifies repetitive tasks and enhances security by rationalizing system administration.
The installation of the Bull system has involved running a large number of benchmarks, to evaluate the performance and scalability of the system for parallel processing. These include the five most relevant benchmarks for materials science (CMT) based on Gaussian software, and a bio-medical and life sciences benchmark (CBV) based on Amber (a molecular simulation program).
The results achieved were 163% higher than those obtained on the hardware installed in 2006.
In summary, the result of the tests carried out clearly demonstrate the excellent performance of the Bull supercomputer in CESCA’s operating environment, which involves moderately parallel tasks requiring a great deal of memory and disk capacity.