The Navy DoD Supercomputing Resource Center (Navy DSRC) is a premier provider of high performance computing services and support to the U.S. Department of Defense scientists and engineers. It is one of five supercomputing centers established under the auspices of the Department of Defense High Performance Computing Modernization Program (HPCMP).
The Navy DSRC leverages high performance computers and a wealth of knowledge to greatly improve the computational research environment for DoD researchers and the computing power necessary to run numerical environmental models for Navy operations. Modeling supports several mission areas including mine warfare, anti-submarine warfare, expeditionary warfare and Navy Special Warfare. Modeling data assists in characterizing and predicting the physical environment which is used by warfighters to successfully execute operations. The center also provides leadership for HPCMP-wide functions including remote petascale data storage, enterprise systems monitoring, and remote system and database administration.
The DoD HPCMP announced in February that it has just completed its fiscal year 2013 investment in supercomputing capability supporting the DoD science, engineering, test and acquisition communities. The total acquisition is valued at $50 million, including acquisition of multiple supercomputing systems and hardware, as well as software maintenance services. At nearly three petaflops of computing capability, the acquisition constitutes a more than 50 percent increase in the DoD HPCMP’s current peak computing capability.
The purchase includes three systems that will collectively provide more than 114,000 cores, more than 307 terabytes of memory, and a total disk storage capacity of almost nine petabytes.
The Naval Meteorology and Oceanography Command (NAVMETOCCOM) operates the Navy DSRC, one of the five DoD supercomputer centers, on behalf of the DoD HPCMP. Located at Stennis Space Center, Miss., the Navy DSRC will receive two Cray XC30 systems built upon the 2.7 GHz Intel Xeon E5-2697 v2 processor and the Intel Xeon Phi Coprocessor 5120D. These two systems are identical, each consisting of 29,304 compute cores, 7,440 coprocessor cores, and 78 terabytes of memory. The systems are designed as sister systems to provide continuous service during maintenance outages. Cray will begin installing the systems in mid-May.
CHIPS asked Bryan Comstock to talk about the new systems and the Navy DSRC March 7.
Q: Are the capabilities in the two Cray XC30 systems a big leap from what you worked with before? How much of a difference will they make to your work?
A: The 1,500 teraflops provided by the new supercomputers will more than double the Center’s aggregate computational capability. We currently have systems capable of about 954 teraflops. Our current systems are IBM iDataPlex systems based on 2.6 GHz Intel Xeon E5-2670 processors and Intel Xeon Phi 5110P coprocessors. Once the new Cray systems are operational, the aggregate computational capability of the Center’s systems will be 2,400 teraflops or 2.4 petaflops.
Navy DSRC computing power doubles every two to three years, typically following Moore’s Law. Our systems have a four-year lifespan. The supercomputers we have now were installed in 2012, and we’ll have them through 2016. The new Cray systems will remain in production at the Center until 2017. At the end of the four-year cycle, these systems are made available to other organizations within the DoD for re-use.
Q: Can you describe the type of work you will be doing? Will you be conducting research or supporting current operations?
A: The Navy DSRC provides DoD scientists and engineers with the resources necessary to solve the most demanding computational problems. The supercomputers enable the DoD science and research communities to test and model defense systems that cannot be modeled in the real-world due to time, financial, physical or safety constraints. In some cases, they can accomplish this work in a matter of hours as opposed to the days, weeks, or even months that traditional research methods can require.
Our systems are shared resources with many DoD organizations using them. The computational systems are divided up into prioritized queues, ensuring that the users share the systems as fairly as possible. With the new systems, our users will be able to scale their work to much larger models and get their work done much quicker.
Shortly after the new supercomputers are installed, we will allow a number of pioneer users to have access to the system for the purpose of scaling their codes on a new architecture. They’ll be able to increase the core counts of typical jobs they might run today, as well as increase problem sizes. The new computing power is going to have a big impact on our users.
Also of note, within the HPCMP, the Navy DSRC is unique in providing supercomputing resources available 24/7 to the NAVMETOCCOM. These high performance computing resources are used by the Naval Oceanographic Office and the Fleet Numerical Meteorology and Oceanography Center for ocean and weather forecasts in support of U.S. Navy fleet operations.
Q: So, for example, does the information the Navy DRSC provides end up in NAVGEM, the Navy Global Environmental Model? Can you talk about the products or services you provide directly to the warfighter and fleet?
A: The Center provides the engines that execute a number of various high resolution numerical models of the ocean and atmosphere. The actual running of these models is managed remotely by the Naval Oceanographic Office and by the Fleet Numerical Meteorology and Oceanography Center. These two organizations, using their own systems, are also responsible for pre-processing data for model input and for post processing the model output and for creating and making available specific products to their fleet customers.
Q: Can you talk about the other organizations that you work with? It sounds like the Navy DRSC supercomputers are in high demand. How do you decide who uses the Navy DRSC, how much time they will have and how much computing power they will be given?
A: This year the Center has over 1,100 user accounts and provides computational support to over 200 projects. The users are located around the country in more than 200 locations. There is a long established process for the scientists and engineers to gain access to HPCMP computational resources, which includes the Navy DSRC. The DoD HPCMP provides the Army, Navy, Air Force, and Defense Agencies, collectively referred to as Service/Agency, with allocations each fiscal year. It is then up to each Service/Agency to determine which projects receive an allocation and on which systems. In addition the HPCMP also supports a few very large computational projects which are peer reviewed. These projects are referred to as either Frontier or Challenge projects.
Q. Do you have any other success stories to share?
A: Within the DoD, HPC is used in a remarkably broad array of activities. HPC has proven to be an essential part of the DoD's ability to design, test, produce, and sustain DoD weapons systems at reduced cost, time and risk with higher performance. The technical report "2012 High Performance Computing Modernization Program Contributions to DoD Mission Success" provides a variety of papers that highlight the research that is being accomplished using HPCMP resources. It is available on the DoD HPCMP website.
Previous editions of the DoD HPCMP Contributions to Mission Success publication have included examples of the use of high performance computing to better enable research and development pertaining to the Navy Seawolf submarine, the Army M1 tank, the Air Force F-22, theater missile defense, and many other key DoD programs.