When the Texas Advanced Computing Center (TACC) was seeking to name an unusual new machine to join Longhorn, Lonestar, and a collection of servers named for things Texan, Maverick came to mind rather naturally. .A Maverick, in the most charitable dictionary definition, is someone unusual, someone who doesn't conform or follow the crowd. The word has its origin in Texas, where the colorful pioneer Samuel Augustus Maverick (1803-1870) allowed his cattle to roam freely on the range. Because he wouldn't brand them, an unbranded calf became known as a maverick.

TACC and Sun Microsystems have collaborated to design, configure, and install Maverick, a supercomputer built to facilitate large-scale data analysis and remote terascale visualization. Dedicated at TACC on October 1, 2004, Maverick is a unique resource for both TACC and the National Science Foundation (NSF) TeraGrid, in which TACC is a partner, because of its excellent balance among large memory, capable computing, and powerful, flexible graphics. "There is no other machine like it in the world," says Dr. Kelly Gaither, Associate Director of TACC, who worked to configure the system, "so the name is really appropriate."

"We are excited to install this leading-edge system for researchers at UT Austin and across the nation. Maverick has the capabilities needed to understand the data resulting from large-scale experiments and simulations rapidly and effectively, thus increasing scientists' ability to address important problems that have urgent or real-time solution requirements," says Dr. Jay Boisseau, Director of TACC. "Collaborating with leading, innovative organizations like TACC to develop state-of-the-art solutions that can later be deployed in more general compute environments is at the heart of what Sun does," says David Yen, executive vice president of the scalable systems group at Sun Microsystems

Maverick is a Sun Fire E25K server for high-end computation, rendering, and analysis. It has 512 gigabytes (GB) of shared memory. The 128 UltraSPARC IV multithreaded processors deliver a peak performance speed of 275 gigaflops, and these are joined by 16 dual-video-out commodity graphics processors (cards), all of which will access a 50-terabyte SAN over eight 2-Gbps fiber channels. The front end for Maverick is a 16-processor Sun Fire V890 server.

"The inclusion of the commodity graphics cards in a modularized fashion makes Maverick special," Gaither says, "because the E25K can be easily upgraded to include next-generation hardware." The graphics cards include a high degree of on-chip parallelism, and the needs of the video gaming industry are powering their rapid development. Each card can drive two displays, allowing either a multi-user environment or visualization on large displays (or, with multiple cards, both).

"The main point is that Maverick is the most capable machine for visualizing large-scale data at interactive frame rates," says Gaither, who directs the TACC Scientific Visualization group. "It has the bandwidth we need to make movies of physical, chemical, biological, and geological processes and to steer models of those processes in real time," she says.

The problems that Maverick can tackle best are those for which the solution is best appreciated through visualization and for which the time to solution is important. These include a variety of real-world problems, many of which come under the heading of emergency management. First responders want to know what is happening, where it is happening, at what level of severity, and what can be done by cooperating agencies to reduce loss of life, injuries, and damage to property.

"We certainly have time-to-solution issues that Maverick will be ideal for solving," says Dr. Gordon Wells. He leads a partnership of academic institutions with links to emergency management entities that are now poised to take a big step forward in flood prediction with the arrival of Maverick.

At the Mid-American Geospatial Information Center (MAGIC), at UT Austin's Center for Space Research, Wells and his team engage in a daily struggle to convert large-scale streams of data into timely information for state and federal emergency management agencies, regional and local governments, academic institutions, TV and radio broadcasters, and the public.

The MAGIC satellite receiving station collects direct transmission of remotely sensed data from orbiting satellites and combines these data with geographic and demographic information to produce geospatial analyses deliverable within minutes in the case of such life-threatening emergencies as wildfires, tornadoes, flash floods, and hurricanes.

"We have a group of partners who want to answer the problem of providing real-time flood hazard forecasts for the decision makers in emergency operations centers," Wells says. In addition to CSR and TACC, the partners include the Center for Research in Water Resources (CRWR) at UT Austin and participants at two other TeraGrid sites: Oak Ridge National Laboratory (ORNL) and Purdue University.

Using scanning laser (LiDAR) elevation data to record surface features within watersheds and NEXRAD radar rainfall data, predictions of the timing and extent of inundation will be made by a model called Map2Map developed by Dr. David Maidment at CRWR. ORNL will contribute dynamic population data--daily movement in and out of cities.

Maverick will be able to predict inundation patterns for specific hurricane and storm tracks, enabling rapid planning and changes in evacuation routes and shelter locations. "What is best about this project," Wells says, "is that it holds the promise of moving quickly from an academic study into an operational mode--in which the loss of life and property caused by storm surge and flash floods could be greatly diminished."

Another project with time-to-solution imperatives is Linked Environments for Atmospheric Discovery (LEAD), an ambitious NSF Information Technology Research project involving nine academic and research institutions, led by meteorologist Dr. Kelvin K. Droegemeier of the University of Oklahoma. LEAD's mission is to provide the information technology infrastructure needed for people (researchers, graduate and undergraduate students, teachers and students in grades 6-12) and tools (meteorological models, analysis systems, data mining engines, and remote sensors such as radars) to interact with weather as it evolves.

In practical terms, Droegemeier expects LEAD to dramatically improve the capability to understand hazardous local weather, such as severe storms, tornadoes, and hurricanes, thus increasing the timely warning of such events. "We are developing the dynamic computing and networking infrastructure required for on-demand detection, simulation, and prediction of high-impact local weather such as thunderstorms," he says.

With LEAD, a user will be able to locate and gather data, process them, and feed them into a computer forecast system that changes its configuration, in real time, both in response to the weather and to available computing resources. Such models create hundreds of terabytes of output that must be mined quickly as their contents determine the next steps in the evolving forecast process. "Far too often visualization isn't though of as a high-end need or resource," Droegemeier says, "and Maverick will change that notion, especially for those of us who work remotely from it. I'm very excited about Maverick and working with our colleagues at UT!"

A number of similar, real-time, on-demand computing projects are slated for investigation on TeraGrid resources, projects ranging from brain surgery to the dispersion of pollutants or diseases in urban areas. "We think all of these projects will find that Maverick can be a powerful resource for modeling and displaying the problems and the solutions," Gaither says.