The blade server cluster solution of Beihang

1. Overall design of the project

A few days ago, based on http://www.aliyun.com/zixun/aggregation/13737.html "> Blade server Shuguang high-performance Cluster to create success, successfully won the bid of Beihang University high-performance Computer Group teaching and research equipment procurement project." The core of this high-performance cluster is a TC2600 blade server designed for trillions supercomputer, equipped with AMD four core "Barcelona" processor, the completed system will provide a theoretical peak of up to 3.9 trillion times per second of the strong floating-point operation capacity, used to carry the school's super large-scale physical and chemical calculation. In this project, the Shuguang company will be the blade server and the newly listed "Barcelona" combined application, fully demonstrated the Shuguang Company's technology advancement.

The whole system is made up of the TC2600 Blade server which is designed by Shuguang company for Trillions, and combines Dawn TC4000 series cluster system technology. The compute node employs a total of 128 1.9GHz four core AMD Opteron processors, "Barcelona (Barcalona)", which provides 512 parallel computing units for users, which can carry large scale physical and chemical computations, with a total theoretical peak floating-point operation capacity of up to 3.9 trillion times per second.

The computing node uses AMD's newest four core "Barcelona" processor, which is the industry's first truly four core X86 CPU, with floating-point performance more than 40% per cent higher than competitor Intel, and is ideal for building large-scale cluster systems, with the current China trillions super computer Dawn 5000, Cray for Oak Ridge The petajoules computer, built by the National Laboratory, and the Petajoules supercomputer (which will contain 16000 cell and 16000 AMD processors), which IBM built for the National Nuclear Safety Administration of the U.S. Department of Energy, will be built using the AMD four core Barcelona processor.

The storage system uses a DS2412FA disk array with dual redundancy hot-swappable RAID controller to improve system reliability and to configure 3TB storage space to meet the storage requirements for large scale computations. The disk array in a single cabinet can support the expansion of up to 18TB of storage space, with excellent price/performance.

The network system uses the global HPC field leading manufacturer Force10 the gigabit switch as the computation net, the S50 switch has up to 192G the back plate exchange bandwidth and the 101MPPS packet forwarding rate, may realize the entire line speed non-blocking data exchange, enhances the parallel computation efficiency. The same is the blade chassis in the Gigabit network switching module with a gigabit switch can be used as a management network, dual-network separation, mutual backup, improve system reliability.

With the use of Blade server to build HPC system, the system power consumption has been greatly reduced. The power consumption of each quad-core Barcelona processor is only 68W, so the power consumption of each blade at full speed is about 240W, while the blade with four core clovertown power consumption is as high as 320W, saving 33%! The total power consumption of the entire system is no more than 17KW, compared with Intel energy-saving 5KW, each year for the national Energy Saving 43,800-kilowatt.

2, TC2600 Blade server

TC2600 Blade Server is the dawning company's third generation of dual core/four core blade server products, it in many ways beyond the traditional rack-type server, is a high stability, high availability, high expansion, high-density, high-performance new generation of server products. It drives the server toward higher computing density, more integration capabilities, and stronger management capabilities, delivering more outstanding computing processing performance, simple deployment usage, and comprehensive monitoring management, which can significantly improve productivity and productivity in data centers and High-performance computing centres, Minimize IT operations and manage maintenance costs. It uses the overall optimization strategy, effectively reduces the system power consumption, coupled with its humanized design, super system expansion capability, excellent cost performance and computational density, can cover a wider range of user needs. Compared to traditional servers, TC2600 blades provide a more competitive overall solution, ideal for meeting higher computing performance requirements and running critical applications.

Prior to this, the University of Aeronautics and Astronautics has been in the automation, science, Aviation Institute and many other institutions used the Shuguang series of products, and are used in the field of scientific computing, for the University of Aeronautics and astronautics to improve the level of scientific research has laid a solid foundation for information. This time, the performance of the "Barcelona" processor and blade combined again applied to the University of Aeronautics and Astronautics, more shows the Shuguang cluster products very high availability, for the scientific research of Beihang University to make a contribution.