| |
The three-chassis Cray XD1™ arrived at ARSC on May 9.
| |
Room to grow: Nelchina is
expandable to accomodate user demand.
|
|
Introducing Nelchina, Cray XD1™
This summer, the Arctic Region Supercomputing Center introduced its newest machine, a Cray XD1™ named Nelchina, to the UAF community.
Nelchina is a three-chassis system with each chassis consisting of 12 AMD Opteron™ processors arranged in six nodes. Each chassis communicates through a high-bandwidth, low-latency, RapidArray interconnection network, which directly connects processors and memory, avoiding PCI bus bottlenecks and decreasing shared resource contention.
The highlight feature of Nelchina is the six Field Programmable Gate Arrays (FPGAs), which are optimized for parallel processing, in one of the chassis. The FPGAs accelerate computationally-intensive segments of code algorithms, which allows the FPGAs to act as co-processors to the Opteron™ processors, significantly increasing system performance. Since the FPGAs do not have fixed architecture, the logic in the FPGA can be customized and optimized to perform only the required task.
Professor Tarek El-Ghazawi, of The George Washington University (GWU), along with five of his graduate students, spent the summer at ARSC and UAF evaluating and configuring the XD1™.
Two of the students focused primarily on implementation and evaluation of the XD1™’s FPGA capability. For some computational tasks, FPGAs are significantly faster than the central processing unit. Implementing an FPGA code can be challenging, though, and there are overhead and bandwidth issues in determining which sections of a particular program should run in the FPGA, and which should run in the CPU. The XD1™’s high-speed interconnect and its model for sharing memory between the CPU and the FPGA module provided the GWU team with opportunities to explore FPGA performance in a variety of application areas, including image registration for synthetic aperture radar and genetic pattern recognition.
The other three GWU students performed studies and analysis of parallel programming languages, with an emphasis on Unified Parallel C (UPC). UPC is a programming language based on C that enables explicit parallelism among programming units. It can offer decreased development time for programmers, as well as faster execution time for the code. Evaluation of UPC took place on several of ARSC’s systems including the Cray XD1™, the Cray X1™ and the IBM Power 4™ p690+/655+.
FPGA Symposium
To understand and investigate the capabilities of reconfigurable computing and FPGA technologies, ARSC hosted a two-day symposium followed by a one-day hands-on workshop. The symposium provided opportunities for discussion and information exchange about the newest innovations in reconfigurable computing technology.
Symposium participants included industry developers from Cray, SRC Computers, Inc. and SGI, as well as Department of Defense researchers and users and researchers from academic institutions. During three days, many potentials of reconfigurable computing were examined through presentations
and workshops.
[Back to Top] |
 |
Frank Williams
ARSC Director
|
|
From the Director
One of ARSC’s primary goals is to enable our users by providing the right tools for their jobs. This means ensuring that we have a variety of systems and access paths available for users to select from as they perform each phase of their computational research.
One of the first steps in this process is advantaging those users, at UAF and beyond, who need quick and easy access to high-end computing resources on a smaller scale than those needed by our larger, on-going research projects. We have recently changed the configuration on our Cray X1™, Klondike, such that the number of nodes reserved for interactive use is increased from one to three during the daytime hours (8 a.m. to 5 p.m. Alaska time). This allows short jobs using up to 48 processors to execute without using the batch queues at all, providing support for user jobs that need more flexible access to our systems during development and testing.
We’ve also been at work over the last year configuring several cluster computing resources, both commodity and custom, and making these resources available to our users. Today, we are operating four cluster systems with a variety of processors and interconnects.
We have also made our XD1™ system available for the ongoing weather forecasting needs of the state of Alaska. Like many remote areas, Alaska is in need of accurate weather prediction for the safety of its residents and visitors. By dedicating nodes during prescribed time periods to this effort, we are able to support experiments in weather forecasting that will eventually provide Alaska residents with reliable forecasting for their everyday lives and business needs. Hopefully, this experience in scheduled forecasting will help us extend this functionality to other ARSC systems and other disciplines.
These kinds of small changes take work behind the scenes, but that effort helps us to meet ARSC’s goal to remain agile while continuing to provide the best computational resources to support the needs of Alaska and the United States. [Back to Top]
|
|
| |
Faculty Camp members immersed themselves in comprehensive supercomputing training for two weeks this summer.
|
Faculty Camp Offers Core Skills Training
ARSC’s three-week Faculty Camp concluded on August 19. Seven faculty members took advantage of the course, which was designed to make supercomputing accessible to research faculty.
A series of seminars and hands-on training, presented by ARSC staff and current users, provided participants with assistance and expertise to focus on independent work and self-guided study for individual projects. The course covered a thorough range of topics: batch processing, scripting, parallel programming concepts and techniques, and basic familiarization with several visualization packages.
Dr. Sveta Berezovskaya studies arctic hydrology with the UAF Water and Environmental Research Center. She used the two-week course to improve her skills while working on her current study of the Kuparuk River Basin. Berezovskaya met her goal to develop increased skills in visualization software and Fortran program optimization techniques. She was impressed with the wide array of techniques presented in the camp.
“The friendly, open and very well qualified staff was great to work with,” she says. “It was also interesting to meet new people from various areas of science and to learn new things and share my research.”
Other faculty members who took advantage of the training include Cheng-Fu Chen from the Department of Mechanical Engineering; Rudi Gens from the Alaska Satellite Facility at the Geophysical Institute; Hideaki Kitauchi from the Japan Marine Science Technology Center; Young-Han Shin from the University of Pennsylvania and Martin Stuefer from the Geophysical Institute.
Additional Training for Users
To help familiarize UAF users with MATLAB®, a versatile high-level computing language, ARSC hosted a one-day training session taught by Khagendra Gupta, a certified trainer from Mathworks, the company behind MATLAB®. The session, held during ARSC’s Faculty Camp, was open to ARSC users and the UAF community.
The day-long session covered many aspects of this very useful pre- and post-processing tool. Topics included plotting and visualization of data and using MATLAB® variables and functions for integrating MATLAB®-based algorithms with external applications and languages such as C, C++, Fortran, Java, COM and Microsoft Excel.
MATLAB® can be used in a variety of ARSC environments. It is easily used on the new Cray XD1™, Nelchina; the IBM Power4™ p690+/p655+, Iceberg; the IBM p690+ Regatta™, Iceflyer; multiple Linux Workstations and the Macintosh systems in the ARSC classroom.
Users quickly filled the 24 available seats and more people lined up in the hallway. With continued interest in MATLAB® training, ARSC has scheduled a second session for November 3 through 5, 2005. ARSC will continue to provide access to technical and training resources in the future. [Back
to Top]
|
|
Three-dimensional visualizations can help scientists translate numerical data sets into expressive information.
|
|
ARSC Specialist Presents at SIGGRAPH
Dr. Roger Edberg, ARSC Visualization Specialist, presented a new series of tsunami animations this August at SIGGRAPH, the international computer graphics conference in Los Angeles, California. The animations employed an innovative combination of scientific programming and art and animation technologies to visualize data from tsunami models created by university researchers.
Dr. Zygmunt Kowalik, of the UAF Institute of Marine Science, and Elena Suleimani, from the Geophysical Institute, each developed a model to understand different aspects of the origin and dynamics of the December 26, 2004, Indian Ocean tsunami. Suleimani also worked on a model of submarine landslides and the resulting tsunami in Seward, Alaska that were triggered by the 1964 earthquake.
By visualizing the modeled data, researchers were able to see the amplitudes and travel times of the tsunami waves, as well as to see where the wave energy was concentrated during the tsunami propagation.
Edberg used an innovative approach combining custom C++ mesh code with movie-industry CGI technology. To achieve the results he wanted, Edberg utilized Pixar’s RenderMan® to create imagery from the high-resolution tsunami simulation data.
His work earned him a lecture spot to describe his workflow and new approach to creating scientific animations. He also discussed the increased efficiency achieved by adapting cinema-quality software to his visualization projects. Though this innovative process requires substantial programming, it proved to be faster and produce higher-quality results than some traditional scientific animation strategies. [Back to Top]
|
|
Alaska Summer Research Academy Cadets Experience Virtual Summer
Eight teenagers took part in the Virtual Reality Programming course taught by UAF Assistant Professor of Computer Science, Glenn Chappell, for the UAF Alaska Summer Research Academy (ASRA). ASRA is a residential outreach program for students entering grades eight through 12. In a two-week course, academy cadets were introduced to graphics programming and progressed to the point at which they created their own interactive, computer-generated, three-dimensional virtual worlds. After writing programs using the C++ programming language, the OpenGL graphics library, and the VRjuggler virtual reality framework on desktop computers, students ran their programs on the visualization clusters in the ARSC Discovery Lab. There, their programs were given life in a virtual reality theater consisting of four 10’ x 8’ display screens that surround the user in a computer-generated world.
Thirteen-year-old Zach Fitzgerald took satisfaction from writing code each day, designing virtual reality programs the first week, and running them in the Discovery Lab the second week.
“Getting to test a program is a highlight,” said Fitzgerald, “and when you’re the person in the ‘cave’ the world is drawn to your perspective; it’s like a movie.”
At first, many of the students weren’t sure how they were ever going to learn each new step; the course condensed several years of computer programming into a matter of days. But by the end of each day, they had mastered a new element of the programming structure.
This was the first experience with programming for 15-year-old Ben Nye. After being introduced to the topic and shown the techniques with which to approach the project, Nye worked with his team to create Particle, a virtual reality game wherein the user wields a spear to destroy attacking particles of three-dimensional fluff.
“Most people here would agree that one of the most satisfying things is having your program compile correctly,” said Nye, “which it didn’t do often.”
“But sometimes,” added Parker Gallagher, 14, “the glitches look really cool.”
In all, the students stayed on campus, experienced the fun of life in an academic environment away from home and away from parents. Most agreed that they wanted to pursue computer science as an area of study in the future.
[Back to Top]
|
|
