HPC Infrastructures Increase in Academia
Teraflop supercomputers, terabyte storage systems and gibabit networks all support academic research.
High-performance computing (HPC), networking and computer storage capabilities in U.S. academia have increased substantially over the past seven years, according to a recent report by the National Science Foundation, Computing and Networking Capacity Increases at Academic Research Institutions, NSF 13-329. These information processing components (HPC, networking and storage) are integrally connected with each other and have become essential for the growth of research across all academic disciplines and also in government and industrial research areas as well.
Academia has continued to gain greater access to high-speed networks through an accelerating high-speed bandwidth of national and regional providers, which includes Internet2, the university-owned LambdaRail and the Dept. of Energy’s Energy Sciences Network. In 2011, 59% of academic institutions stated they had at least 1 Gb/sec, up from just 21% in 2005. Those universities having at least 10 Gb/sec network connections increased from 2% in 2005 to 25% in 2011.
Nearly 200 of the NSF’s 539 surveyed academic institutions also reported that they maintained HPC resources of 1 teraflop or faster in 2011. Indeed, of the Top500 (www.top500.org) fastest supercomputers in the world (June 2013 ranking), at least 90 are listed as being in academia with 21 of those being in U.S. academia. Some of the sites in this list are noted as “research sites,” which physically can be on university campuses as well. The last academic site on the list, No. 497, is Clemson Univ., S.C., with a Dell/Sun/IBM system with a performance rating of 96.9 Tflop/sec (Rmax).
The NSF report noted that in 2011, 24 of the 539 surveyed institutions had combined computing capacities of at least 100 Tflops. Of course, this is not a fixed level and researchers continue to upgrade their systems, so the NSF level has most assuredly increased since then.
As academic research increases, the collection of massive data sets has similarly increased, which requires the implementation of increasingly massive computer-based physical storage capabilities along with appropriate data management systems. Indeed, data management plans are often required in research grant proposals where large data set will be used.
Of the academic institutions surveyed by the NSF who maintained centrally administered HPC resources in 2011, more than half (56%) reported online storage capabilities of more than 500 TB. A smaller share of public (21%) and private institutions (18%) reported they provided greater than 500 TB of online storage for their users.
The development of high performance computing systems in the U.S. is driven by a few major mostly U.S.-based vendors (IBM, Cray, Hewlett-Packard, Intel, NVIDIA and Dell), long-term development programs at the U.S. Departments of Defense and Energy, and basic research programs in academia. This has maintained U.S. dominance in this technology arena. And while about only half of the largest supercomputers in the world are sited in the U.S., more than 90% of all HPC systems are supplied by U.S. vendors, according to the latest study by industry analyst, IDC. That does not mean the U.S. has a lock on HPC systems though, as more nations are developing their own HPC hardware and software. To see the ready challengers, you only have to look at the top HPC system in the world, the Tianhe-2 developed by the National Univ. of Defense Technology, which has nearly twice the performance of the number two system—the Cray-powered Titan at the DOE’s Oak Ridge National Laboratory. And along with that trend is the related trend of U.S. vendors adopting the UK-based ARM computer designs.
Government officials increasingly recognize the value of HPC systems for their country’s economic competitiveness, as well as for enhancing the development of science. HPC systems are too strategic to outsource to other countries. This is what drove China to develop its own Tianhe-2 system (along with Sunway and Godson systems), Japan to develop Fujitsu SPARC-based systems and Russia to develop their T-Platforms. The economic importance of HPC has a growing recognition that will boost future indigenous technology development, much of it in U.S. academic settings.