2005 Computerworld Honors Program

	Scalable Concurrent Programming Lab's advantage in micro-electronics, satellite manufacturing and launch vehicle design California Institute of Technology Pasadena, CA USA Year: 1996 Status: Finalist Category: Science Nominating Company: Cray Research, Inc.
	Parallel supercomputers help lower the costs of space operations by simulating the behavior of satellite launch vehicles, allowing flaws to be found in systems that cannot be tested on the ground because they create too much heat.
	Satellites now provide us with a broad range of services including long distance telephones, television, modern navigation used by commercial airlines and ships, television, and weather surveillance. Launch vehicles, such as the Titan IV and Delta II, are the U.S.A.'s primary method for placing a satellite into orbit around the earth. Due to the temperatures involved, it is not possible to fully test a launch vehicle and its boosters on the ground. If a problem occurs on a flight, there is little information available to allow engineers to isolate the problem and suggest a fix. An alternative is to recreate the flight on a computer using the basic laws of physics and aerodynamics. Unfortunately, this is a complex, costly, and very time-consuming task. As a result, until recently simulations have had little impact on real missions. A capability has recently been developed that allows the flight of a launch vehicle, with its boosters burning, to be simulated using not one computer, but hundreds and potentially thousands of computers. The computers are connected together using new information technology that allows them to exchange information and cooperate to achieve a common goal. In this case, the computers work together to create a single flight simulation in a fraction of the normal time: A simulation that might take months on a single computer, can execute in a matter of hours on 500 computers. The simulations can be of direct use to engineers who are attempting to resolve flight problems under the time constraints of a mission. The capability was recently used in the investigation of an anomaly that occurred on the flight of a Delta II vehicle. This anomaly caused the release of a satellite into the wrong orbit. A set of simulations were produced in less than two weeks that showed how the heat from the boosters affected the flight of the vehicle. Simulations of the flight with the boosters on and with them off were produced to allow comparisons with recorded flight data. This information was used by the investigating team to understand and resolve the issue for future flights. The simulations were conducted using 512 computers coupled together in machine called the Cray T3D. This machine uses fast switches to allow the computers to exchange information. The simulation capability can operate on a broad range computers and the computers can be connected using a wide variety of networking technologies. The capability has already had direct impact on our industrial competativeness in the satellite launch business. As information technology improves, the capability will allow increasing large number of computers to be brought to bear on problems associated with vehicle flights. This increased computing power not only allows faster simulations, but also more accurate simulations. Thus we expect that in the future the capability can provide direct improvements in both the cost and safety of space travel.
	The capability has already had direct impact on our industrial competativeness in the satellite launch business. The capability was recently used in the investigation of an anomaly that occurred on the flight of a Delta II vehicle. This anomaly caused the release of a satellite into the wrong orbit. A set of simulations were produced in less than two weeks that showed how the heat from the boosters affected the flight of the vehicle. Simulations of the flight with the boosters on and with them off were produced to allow comparisons with recorded flight data. This information was used by the investigating team to understand and resolve the issue for future flights. The simulations were conducted using 512 computers coupled together in machine called the Cray T3D. As information technology improves, the capability will allow increasing large numbers of computers to be brought to bear on problems associated with launch vehicle flights. This increased computing power not only allows faster simulations, but also more accurate simulations. Thus we expect that in the future the capability can provide not simply an aid to fixing problems but a tool for designing new vehicles. This can provide direct improvements in both the cost and safety of space travel.
	The capability leverages two new information technologies: 1) SCPlib -- a new software technology developed at Caltech for programming large numbers of computers and 2) new scalable parallel computer technology -- machines that are able to grow by adding more and more computers to them with advanced networking technology. The former provides an efficient method for the design of new programs that operate on potentially thousands of computers. This technology is capable of automatically redistributing work among the computers while the program is executing so as to increase performance. It also allows the program to execute on many different types of parallel computer. These new computers include multicomputers like the Intel Paragon or Cray T3D, shared memory multiprocessors such as the Silicon Graphics Power Challenge, and a broad variety of networked workstations such as the IBM RS6000, Sun Sparcstation, Hewlett Packard 700 series, and Silicon Graphics Indego family. To cope with the differences in these machines, programs must incorporate methods to trade communication on the network for additional work to be executed on a computer.
	The capability is unique in both its scaling and portability aspects and utilizes state-of-the-art parallel computing techniques. We are not aware of other capabilities that have been used successfully for simulations of launch vehicles under mission critical time-scales. The numerical techniques for solution of the physics grew out of a proprietary uniprocessor code that was operated on Vector Supercomputers by The Aerospace Corporation. This original code expressed the physics to be solved but could not operate on parallel machines. It took four years to develop the parallel code, include the physics from the original code, validate it on standard tests, and apply the new parallel code to real launch vehicle problems. This period involved periodic redesign to take advantage of changing technology.
	The application is fully functional and in constant use by scientists at The Aerospace Corporation. It has achieved the goals of the program, namely to create a capability that can scale to utilize thousands of computers and drastically increase the turnaround time for simulations. At the outset, four years ago, the work was considered genuine research in new technology and although potentially viable, was not then expected to directly contribute to missions. The primary beneficiaries are anyone using satellite technology -- without reliable launch technology these devices cannot be maintained in service. In future we plan to develop versions of the capability than can cope with vehicles that separate during flight. This moving boundary type of problem will allow us to broaden the scope of possible simulations that can be performed.
	There were three primary difficulties: 1) Communication between individuals. The team is multidisciplinary in nature leveraging skills in parallel computing, graphics, grid generation, networking, and physics. Each individual speaks a different language from a technical perspective and comes with a different educational background. As a result, it is difficult to express an idea and have that idea understood by other members of the team. Moreover, since every team member is not an expert there is a significant gap in understanding. Much patience, practice, and hard work is required to overcome the prejudices and frustrations of communication so as to couple technologies effectively. 2) Technology Evolution and Design. There have been numerous types of parallel computing and information technology available over the last five years. A pressing task has been to evolve the ideas both and underlying software technology to keep up and take advantage of this changing hardware technology. Today, the capability can operate on virtually every interesting hardware platform available, and can scale to machines that include thousands of computers. However, this has been achieved only through a constant reevaluation of direction in light of changes in the underlying information technology. Software engineering to facilitate this change required a substantial investment in time. 3) Available Machine Resources. Even today there remain only a few places in the world where sufficiently large machines exist to be able to conduct launch vehicle simulations. As time has proceeded the number of large machines has increased, however, so has the number of competing groups who require access. As a result, large blocks of machine hours to carry out complex simulations are still difficult to arrange. Typically it requires many long hours of details setting up networks, disks, usage quotas etc. However, the cost of the basic technology is decreasing every year due to the explosion in commodity microprocessor use. As a result, we expect large simulations to be commonplace in the next century.