Theory of a Large Supercomputer

Published on November 26, 2013

Theory of a Large Supercomputer

    Dell HPC comes to the aid of scientists
    About the British scientists who have presented the world with more than a dozen discoveries of varying degrees of doubt, we have repeatedly heard. But discussing the next of them, we rarely think about how discoveries are made and what is needed to make them happen. We will leave aside the scientific base, experience and training of scientists (this is all a topic for a separate material), and we will concentrate on the technological component of their work. It is logical to assume that in modern universities calculations are performed on powerful modern technology, experiments are carried out on the latest laboratory equipment, otherwise it will be practically impossible to bring something new to science these days.
    So what kind of technology do scientists from world-famous universities use? At what level are computers and servers capable of handling the crazy load - such as a result of which we are witnessing a new discovery? Today we want to talk about our modest contribution to science and the Dell HPC High Performance Computing.

    It all started with the fact that the George Washington University of America had an urgent need to introduce a fundamentally new technology based on a high-performance system that would satisfy all the needs of university employees. The task was initially not easy, because it was about one of the largest centers of scientific research in the world. George Washington University accepts students, teachers and researchers from all over the world, and its leadership does not hide the desire to make the institution one of the best in the United States and, accordingly, to attract the most outstanding specialists into its ranks. For these purposes, the university spares no money and has already invested a lot in network infrastructure and data storage. But, as you might guess, this is just a drop in the bucket:

    To make it easier to understand the scale of the calculations that are carried out by university researchers, try to imagine a thousand million equations with a thousand million unknowns. To cope with this kind of tasks can only a very powerful system. A high-performance computer was needed not only by physicists, but also mathematicians, biologists, chemists, engineers - any exact science in our time cannot move forward without reliable technological support.

    However, one aspiration was clearly not enough. There was simply no place at the university to establish computing clusters. If the educational institution was located outside the city, then this problem would not have been, but in urban conditions, every square meter was counted. The companies that participated in the tender proposed to split the system into several parts and install small clusters in different buildings, but such a solution did not suit the management, since it made the administration of such a system much more time-consuming and time-consuming. Thus, the university needed a centralized system that could serve several faculties at the same time.

    And it is not known what would have ended if the university management had not turned to Dell in time. The Dell team discussed their needs in detail with scientists, and as a result, the Colonial One HPC cluster was installed in the science and technology building. It consists of 96 nodes, based on Intel Xeon E5-2670 processors. The cluster is connected by the FDR InfiniBand bus, which provides data transfer rates of up to 54.5 Gb / s on each port. It integrates all cluster nodes with two storage platforms. The solution is based on Dell PowerEdge C8000 Series servers, 4U flexible chassis, GPUs and storage nodes. The latter is covered by the Dell PowerVault MD3260 array, which can accommodate 180 TB of information in each 4U high bay. The cluster also includes the HPC Dell | Terascala storage solution, a 300 TB high-performance file system. Any university community can get centralized access to the cluster.



    By the way, the merit of Dell is not only that we were able to assemble a relatively compact system that fit in one room without sacrificing performance. The fact is that the Colonial One supercomputer was built from scratch in two weeks. Yes, only 14 days.

    Time-lapse video of the whole process from different cameras:


    With the Dell HPC system, George Washington University employees were able to conduct research in a shorter time frame. What used to take weeks, now takes about one to two days - for example, this applies to genetic and microbiological studies. Thanks to Colonial One's ultra-productive solution, it now only takes 24 hours to determine the species or strain from a sample of its DNA, whereas before this procedure took more than a week. Reducing the time during which the nature of microorganisms is determined means a more effective and timely treatment in the future.
    Among other things, the HPC system allowed scientists to conduct research that were previously inaccessible to them due to insufficient computing power of the equipment. The Faculty of Physics of the University, for example, is now exploring new directions in the study of lattice quantum chromodynamics. Without certain calculations that only a powerful computer could provide, they were previously inaccessible.

    The high-performance Colonial One system helps university management attract new researchers to its ranks, as it has virtually no analogues in other educational institutions and research centers. Many promising scientists see new opportunities for their work in the Dell HPC-system - you must admit that few will give up the opportunity to ease their own work and bring another scientific discovery closer.

    Another important advantage of introducing such a solution is the simplification of cooperation between departments of the university. Previously, physicists, biologists, chemists, engineers worked with local servers, now they have the opportunity to use a single centralized system and conduct joint research.
    The Colonial One system will be developed along with the university. As the needs of researchers increase, cluster capacities will increase. The architecture of the Dell C8000 allows you to consistently replace components and expand the system, including increasing the capacity of drives. Perhaps our solution will allow the university to effectively interact with other research organizations.
    It is always nice to make your own contribution to the development of science :)

    PS A little video in the end