The solution of dawning cloud computing topology in some universities in China

In order to improve the quality of scientific research and ensure the successful completion of scientific research work in China, the school hopes to build a school "internal cloud" to unify the scientific research system required by each college in this "internal cloud" system to complete the unified deployment and management.

In order to solve the problem of the actual demand of high performance computing in different disciplines, this high-performance computing system will be used to meet the computational requirements of different numerical science simulations including oil seismic processing, material chemistry, hydrological applications, GIS systems, etc. Therefore, this system for the calculation of the performance of the high requirements for the stability requirements of a very strong, storage requirements, the requirements of the management is extremely mature. With the help of the famous high performance computing solution Provider-Shuguang company and the CAS Computing Institute, we built a high-performance computing platform for building our own "cloud computing".

Excellent computational performance

In order to meet the needs of users for computing performance, the computing subsystem in this scheme has a total of 92 Shuguang cb65-f Blades Server, which is a dual design, using the upgraded version of AMD "Shanghai" processor-six nuclear Istanbul processor, A total of 736 2.6GHz processor cores and 1.5TB of memory are available.

The "Istanbul" processor product is compatible with the existing AMD 4 Core processor interface and is up to 30% performance with the same power design. 6 Core Hao Long processor "Istanbul" employs 45 nm manufacturing process, with 6MB three cache, and integrated dual-channel DDR2 memory controller, compatible socket F interface, power consumption of about 40 watts. AMD "Istanbul" 6 core Hao long processor leverages the existing platform infrastructure and low-cost, energy-efficient DDR-2 memory, helping to reduce system procurement costs. Compared with the previous generation of "Shanghai", the core quantity increases and continues to improve the speed of the bus on the processor.

In addition, the case also provides two fat computing nodes, with strong computing power, theoretical peak operation speed of up to 8.87Tflops (8.87 trillion times per second floating-point operations). Fully meet the requirements of the University for computing performance.
Ultra High Computational density

In today's highly developed society, the cost of space becomes more and more obvious. and Shuguang Company This solution Ultra high computational density became a big bright spot of this project. Shuguang developed a dual-road quad-core and four-way quad-core high density blades with independent intellectual property rights, and a 10-piece industrial standard blade server in China's first 7U height based on the motherboard, which uses the connectx DDR two-layer switching module, the built-in Management Ethernet network switching module, It can realize the ultra-high density of the highest 40cpu,160 core of single box, and realize the OpenMP parallel computation within single node. These high-density computing features effectively save the user's space costs.

Ultra Low Power consumption

In the increasingly serious energy costs today, energy conservation and environmental protection has become the keynote of the overall society. With blade design in this high-performance computing solution, this project can save 20% power consumption than conventional rack-and-cable solutions; it uses liquid cooling system cooling technology, compared to conventional air-conditioning refrigeration 30% air-conditioning power consumption; At the same time, the scheme uses energy-saving software and job scheduling combined, The system is expected to save 20% power consumption throughout the year.

Double-insured water-cooled cooling

The High Performance computing solution for the university users is the first time to adopt a water-cooled system solution for the large-scale cluster system in China, which reduces the requirement of the high performance computing system to the customer room and improves the cooling efficiency. Water cooling system design realizes the separation of water and electricity, and ensures the safety of the system.

At the same time, the water-cooled system also includes the passive cooling system of backup, which can ensure the normal operation of the system under the condition of water cooling failure, and improve the stability of the system.

In short, as a new generation of high-performance computer for national economic construction and social development, Shuguang 5000A is suitable for all kinds of large-scale scientific calculation and engineering calculation, can meet the needs of users, and can maintain a certain technological advance in the next 3-5 years.

Very large storage space

The High-performance computing solution employs the current global shared parallel IO architecture on the industry's leading high-speed InfiniBand interconnect, including 4 data servers, 1 metadata servers, a 12T of one-tier real-time storage and a set of two-level backup storage, The software part is based on the mature lustre parallel file system, the data block is stored in 4 storage management nodes, which reduces the risk of data loss. Combined with Shuguang cluster management software, it provides good support for file system deployment, file system management and rapid recovery of file system.

First-level storage is a high-performance disk storage that is used to hold the original dataset, temporary data, and staging calculations that are required for the calculation during the calculation. In the first tier storage, the dawning enterprise-class disk array ds8314ff is configured with 8 FC host interfaces, equipped with 8GB/S cache, with a maximum expansion to support over 224 disks and direct access to the SAN fibre switch network.

Second-tier storage is used to store data that needs to be kept for a long period of time, such as result data, some public data resources, or other data that the user wishes to keep for the long term. In level two storage, the dawning enterprise-class disk array DS2412FA is configured with 8 FC host interfaces, equipped with GB cache, a maximum expansion of 80 disks, and direct access to the SAN fibre switch network.

Eliminate wasteful resource integration

In this plan, Shuguang company according to the user's request integration entire campus area decentralized supercomputing resources, constructs the region the comprehensive computing platform, does not have any compatibility question and the technical obstacle.

Excellent management system

In this solution, Shuguang company not only provides advanced high-performance computing hardware solutions, at the same time, Shuguang 5000A high-performance Computer also provides a wealth of software systems, including operating systems, compilers, debugger, function library, parallel Communication library, the global sharing of parallel file system and other basic software, As well as the GRIDVIEW2.0 Integrated Monitoring management system and industry-leading PBS Gridworks job management software developed by Shuguang Company for advanced users in the industry, it helps users to manage their High-performance computing centers conveniently and efficiently from the aspects of hardware, system, application, energy consumption, user behavior and so on.