Jilin University undertakes numerous national scientific research projects under the National 863 Program and the National 973 Program. The University has a host of laboratories for theoretical chemistry, auto dynamic simulation, super-hard material and others. The University decided to invest 13 million Yuan to establish a giant high-performance system as the basic platform of its major projects to improve the efficiency of its scientific research. This platform involved large scale and complex applications, with as many as 100 different types of applications spanning more than ten different sectors such as computational chemistry, material, molecular dynamics and CAD/C. As the different applications have diversified features, it is necessary to select different computing software, computing methods and hardware resources according to the different computing tasks.
Analysis of Application Features
The main users of the supercomputer center include the National Key Laboratory of Theoretical Chemistry, the Material College, the National Key Laboratory of Inorganic Synthesis and Preparation Chemistry, the Mathematics College, the Integrated Optoelectronics College, the Automotive College and the National Key Laboratory of SuperMolecular Structure and Material. The applications adopted by various laboratories span various sectors such as Gaussian, ADF, Materials Studio, VASP, Molpro, Molcas, Amber and Grom of chemistry, AMBER, HyperChem, ChemLab and CHARMM of molecular dynamics, and the CAD/CAE’s LS-DYNA, FLUENT and MATLAB.
Based on the application features of software and as well as drawing upon Inspur’s years of clustering experiences, Inspur proposed the following solution:
High-performance computing platform: other than establishing a stable hardware platform, it is also necessary to have a similarly advanced software platform. The completed system shall maintain technical leadership for a certain period of time. It shall satisfy the application scale of the college and the needs for large-scale computation. Mainstream computing software shall run stably and efficiently in the system.
Ease of use: considering the massive research workload of the teacher or instructor, easy operation of the system becomes a key requirement. It is thus necessary to ensure that the instructors can apply the system in a skillful and time-efficient manner.
High scalability: high-performance computation has infinite requirements on computing ability. The completed system shall satisfy certain scalability requirements and its ease of expansion needs to be guaranteed.
Inspur specifically provides tailor-made solution comprising “high-performance computing processor + high-density blade + high-speed network + high-efficiency storage”. This solution fully integrates the latest technologies which include the world’s highest-density HPC blade server NX560T, QDR Infiniband, 8Gb fiber storage and GPU mixed computation. Moreover, all nodes are furnished with energy-efficient power sources with a power conversion ratio of over 94%. A low-carbon high-performance computing system is envisioned on the basis of the system’s green and energy-efficient features.
Upon its successful deployed at the Supercomputer Center of Jilin University, Inspur Tiansuo TS10000 high-performance server’s computing capacity is projected at 40 trillion. [OYC1] This system ranks the first among all high-performance computers of universities nationwide (note: March 2010 statistics) and is ranked among the Top 500 supercomputers worldwide.
The computing module, which comprises “high-density blades + QDR Infiniband”, features the highest computing capacity and a computing network of high bandwidth and low delay. Among the host of user applications, Gaussian and ADF rely heavily on the main frequency of the processor for their computing efficiency; Amber and Grom have very stringent requirements on the network for parallel computation. Therefore, this computing module integrates several latest technologies to eliminate system bottlenecks to the maximum extent. “LUSTRE parallel file system + optical fiber” satisfies the needs for data-intensive computation. The user’s main application system Gaussian will generate big temporary files and therefore file reading and writing will account for half the computation time. The high-speed reading and writing via 8Gb optical fibers directly increases the user’s computing speed and saves time for the user.
Customization of hardware system represents just the first step; the deployment and debugging of application also meet professional standards. During the process of cluster deployment, Inspur engineers and the researchers of the Supercomputer Center of Jilin University worked together for all-around optimization of the applications.
Moreover, the solution also includes Inspur’s self-developed monitoring and management system, which allows all system management work to be performed on the interface. In addition to helping the user to deploy applications, Inspur also utilizes its high-performance training service system to provide comprehensive training to the user in operation and maintenance at the later stage. As such, Inspur not only provides good equipment to the user but also enables the user to utilize the system well.
Upon the supercomputer system’s completion, it will become the platform of scientific research work for all faculties and students of Jilin University. A stable, reliable and easily operable high-performance platform is adopted to establish the cluster and this system features outstanding performances in its ease of use and power consumption.
With the successful deployment of Inspur Tiansuo TS10000 high-performance server at the Supercomputer Center of Jilin University, the system has a computing capacity of 40 trillion[OYC2] . It ranks the first among all high-performance computers of universities nationwide and also ranks among the Top 500 supercomputers in the world.