|
|
|
A significant part of modern scientific research involves the solution of large systems of differential equations that prescribe the state of a system under study. Whether this effort is to understand the complex chemical structure of a compound, simulation of a fluid as it flows around buildings in an urban environment or to understand the dynamical processes associated with mesoscale storm development, the use of sophisticated computing resources is commonplace in academia. Unfortunately, these tools have historically been scarce at the University of North Dakota (UND) and have inhibited the development of competitive research in various computational science fields. While some efforts have been made in the John D. Odegard School of Aerospace Sciences (OSAS) over the past twenty years to bring such computational tools to UND, even these efforts have fallen short of the expectations of UND researchers. Rather, during the past twenty years UND has focused on the use of off campus resources that were coordinated and supported by agencies at other universities or research laboratories. The OSAS Scientific Computing Center was established to bring high-performance computing to UND and OSAS. These efforts included acquisition of several Cray Research supercomputers including the Cray X-MP, Cray Y-MP and the Cray J90. In each case these systems failed to either have the technical and financial support to render them useful to serious computation science or in the case of the J90 the system was acquired with a configuration woefully inadequate and without necessary software tools to make the system usable. Even for researchers who added significant software packages to the system, the overall cost of the system to research was overpriced and not competitive with other high-performance computing resources available to UND researchers at other non-UND facilities. Beyond the efforts at OSAS, no other serious effort has been made at UND in the past quarter century to support serious computation science endeavors. Presently, the UND Regional Weather Information Center (RWIC) and the UND department of atmospheric sciences are engaged in high-performance computing research through a contract with the University of Minnesota's Army High Performance Computing Research Center. This work, while providing personnel and computing resources to support research efforts under this contract, does not provide support for other research opportunities outside the scope of Army research. This limitation is presently a restriction to new research funding development and to the attraction of quality graduate students interested in graduate programs involving computation science. For the department of atmospheric sciences, the lack of UND high-performance computing resources has made the pursuit of advanced atmospheric modeling research unrealistic. The reason for the lack of high-performance computing resources has historically been due to the high cost to acquire and maintain these resources. Past efforts at UND OSAS to maintain Cray computers cost in excess of $250,000 per year including hardware/software maintenance agreements and the needed technical staff to keep the systems running. However, the high-performance computing world has changed dramatically in the past decade. Most notably this change has taken place with the improvement in affordable processor technology that now provides as much computing power as all of the past OSAS Cray computers combined in one desk side workstation. Further, the proliferation of open source operating systems such as Linux and FreeBSD that are available at little to no cost has made the cost of operating research computing systems more attractive than any time since computers have been found in a research environment. Finally, the growth in popularity and advancement in technology involving massively parallel processing have made the construction and availability of high-performance computing in a research laboratory attractive to focused computation science projects and programs. Specifically, the popularity of the Beowulf concept, where high-performance computing environments are constructed from commercial-off-the-shelf (COTS) components, has made affordable scaleable computing a reality. It is this Beowulf concept that will bring high-performance computing to RWIC and provide the capability to engage in advanced weather modeling research. This effort is expected to heighten not only the ability of UND researchers to attract new research funding, but will provide a resource for students to be educated on high-performance computing and participate in leading edge research that will enhance their education at either the graduate or undergraduate level. The UND Regional Weather Information Center will begin construction during February of 2003 a 32-processor Beowulf cluster computer. This system will serve as a base system having over 6 Gigabytes of memory and 1.5 Terabytes of online storage and will be based upon the powerful and industry accepted 2.4 GHz AMD Athelon processor (system specifications is attached to this document). This cluster will eventually be scaled over the next several years to a final size to include over 600 processors with the capability of providing computer resources to address some of the most challenging atmospheric science model problems currently posed today. The Beowulf, with the system name to be referred to as Zeus after the Greek God of Lightning, will be located in the operations room of RWIC and will be eventually housed in up to ten 19-inch computer racks (the initial system will be located in one rack). Special emphasis for the Beowulf cluster will be to address research efforts associated with ensemble atmospheric modeling and finite element modeling of surface transportation simulation of roadway weather conditions. Completion of the first 32-processor Beowulf is expected during April 2003 with use by researchers and students during the spring 2003 semester. The cost to develop the initial 32-processor configuration will be approximately $30,000 including all necessary hardware and software. This cost does not include the cost of technical and scientific staff required to complete the system construction and configuration. These costs are anticipated to be under $10,000. Funding for the initial configuration construction will come almost exclusively from RWIC's local funds, which largely come from indirect cost return from RWIC research endeavors. These funds are appropriate for this funding, as these will reinvest funds to foster new funding opportunities. The cost of personnel to support the system, both construction and operation, will come from existing contracts that will be deriving immediate benefit from the Beowulf's availability. The availability of the RWIC Beowulf to research outside of RWIC will be addressed on an individual case and will be subject to system resources available beyond the needs of RWIC research activities. However, other university researchers and departments will be encouraged to join the RWIC Beowulf development and help distribute the development costs and ongoing operating costs. Beowulf Project Participants John Nordlie, System Administrator, RWIC
For more information, please contact RWIC using the form below.
|
| RWIC Homepage | STWRC | Atmospheric Sciences | Aerospace | University of North Dakota |
Contact Information:
Regional Weather Information Center
PO Box 9007
4125 University Avenue
Grand Forks, ND 58202-9007
Phone (701) 777-2479
Fax (701) 777-3888
webmaster@rwic.und.edu
