 |
|
|
|
|
Company Success Stories - Boeing Space and Communications
Linux NetworX Cluster Supercomputer Boosts Boeing’s Delta IV Rockets
Profile
Boeing Space and Communications
"Cluster Supercomputer Powered by AMD Athlon™ Processors"
Summary
In 2001, a team of engineers at Boeing Space and Communications in Huntington Beach, California, acquired a new AMD Athlon™ processor-based cluster supercomputer from Linux NetworX to perform aerodynamics analyses for a new family of rockets, named Delta IV. The Linux NetworX system, an AMD Athlon processor-based cluster, is being used to perform computational fluid dynamics simulations of the flow environments that the Delta IV rockets will encounter during flight. The computational simulations allow engineers to evaluate various flight environments and their effects on the vehicles prior to actually flying the vehicles, thereby reducing many of the uncertainties associated with each rocket’s flight, and increasing the Delta IV’s reliability.
Overview - The Challenge of Introducing A New Family of Rockets
Boeing Space and Communications (S&C) is the world's largest space and communications company. S&C is home to such programs as the space shuttle and the International Space Station. In 2002, Boeing will begin providing launch services with the Delta IV, a family of launch vehicles resulting from a partnership between The Boeing Company and the United States Air Force as part of the Air Force’s Evolved Expendable Launch Vehicle program.
The Delta IV rockets are configured to accommodate the satellites of the future, which are expected to be larger and heavier. The Delta IV program has been designed to produce, process and launch rockets with increased safety, efficiency, reliability and affordability.
In today’s global launch industry however, there is an overabundance of launch service providers in relation to the number of satellite operators. In order for a launch service provider to stand out, it must offer satellite operators a good price, the right size rocket, vehicle production to meet a customer’s launch schedule, backup launch service options, and, most importantly, a reliable ride to space.
Launch customers typically consider a launch vehicle’s reliability over all other factors. This is the biggest challenge for the new Boeing Delta IV. Any new rocket is unproven until it demonstrates a series of successful launches and can be deemed reliable.
Knowing this, Boeing structured the Delta IV program to reduce the risks associated with the new rockets by addressing Delta IV’s reliability before its initial flights.
AMD processor-based Linux NetworX Supercomputer: New Solution Helps Increase Delta IV’s Reliability
Products like the Delta IV, developed for such customers as the U.S. government, are subjected to very high standards of analyses, testing and verification before being delivered.
In order to help increase Delta IV’s reliability, Boeing engineers are evaluating the rocket’s flight characteristics before the rocket ever leaves the ground in an effort to substantially reduce the risks and uncertainties involved.
With the help of a new AMD processor-based cluster supercomputer being used at the Delta IV’s home in Huntington Beach, engineers are performing computational fluid dynamics (CFD) analyses of the vehicles. These computer-aided simulations provide valuable insight into the interactions between the vehicles and the atmosphere during flight. Similar simulations are also performed to study characteristics of the environmental conditioning system flows inside the vehicle prior to flight.
The CFD simulations provide data to characterize the impact of flight profiles and associated aerodynamic environments on the rocket’s structure and control system. This database helps the engineers to ensure that the rocket will maintain control authority and structural integrity throughout its flight profile for the full range of atmospheric conditions and winds that the vehicle could experience at launch and during flight.
The system was specifically configured by Linux NetworX to meet the Boeing Delta IV aerodynamics team’s computing requirements to run these CFD simulations. The system consists of a host node and 96 compute nodes, each configured with an AMD Athlon processor. The system provides the team with valuable data for accurately understanding the aerodynamic environments in which the Delta IV vehicles must operate. The benefits the AMD Athlon processor-based cluster system brings to the Delta IV program include an efficient platform for performing aerodynamics analyses; exceptional system performance; and very high system reliability.
Thinking Outside the Box
One of the fundamental premises of the Delta IV program is to maintain a sustainable low cost launch service while simultaneously achieving the highest level of reliability. Successfully achieving both stringent cost and performance objectives required a departure from status-quo thinking.
Two historical challenges that the Boeing Delta IV aerodynamics team eliminated by acquiring the cluster system were their reliance on government run supercomputing facilities and their need to purchase costly upgrades to their heritage high-end computing systems to expand capability.
“A number of years ago, when we looked for computing resources to run these simulations, we sought grants to get supercomputing time from NASA and Department of Defense computing labs,” said Ward. “More recently, we purchased large shared-memory systems with many processors. We spent nearly a million dollars to purchase a platform that would become obsolete in three or four years.
“Rather than relying on government labs or the purchase of high-end system upgrades, we chose to invest in a system that would allow us to configure it for our current requirements and allow for affordable expandability in the future. When we evaluated the benefits of the short and long term investment of the new cluster supercomputer versus the uncertainty of obtaining government lab time or the cost of high-end system upgrades, the numbers turned out to be very compelling,” Ward said.
The Cluster Supercomputing Advantage
Cluster supercomputing is a method of linking multiple computers, or nodes, together to form a unified and very powerful system. Cluster supercomputers can be complex to manage and maintain, which is why Boeing Delta IV engineers worked with Linux NetworX on implementation. Linux NetworX provided access to its Supercomputing Center, allowing the Boeing Delta team to optimize the hardware configuration and ensuring its application would run over a cluster. Linux NetworX also provided cluster management tools, which allowed the team to focus on its CFD testing rather than system management.
According to the Delta IV aerodynamics team, the Linux NetworX ClusterWorX software provided a single, user-friendly interface to the large collection of computing nodes and offered great efficiency in operating and maintaining the system.
“The cluster system has one host node that serves as a front-end interface to the 96 individual compute nodes. The single interface makes it possible to efficiently manage the large computing system,” said Ward. “The host node and the compute nodes each run an identical Linux operating system. With this arrangement, a large number of nodes can be efficiently monitored and maintained from a single user-friendly interface, allowing us to spend more time performing aerodynamics analyses, and less time performing system administration. This translates directly into cost savings and improved productivity for the aerodynamics team.
“Another key factor in our selection of a system was that we needed a computing solution that had the capability to run our simulation software without requiring a tremendous amount of software rework. This criterion was of particular concern for the distributed-memory cluster systems, since we had only run our software on shared-memory servers previously. We needed to be certain prior to purchasing a distributed-memory cluster that the system would run our jobs concurrently across its many nodes. As it turns out, our software runs efficiently in a distributed-memory environment,” added Ward.
High Reliability
Reliability of the computing platform was a vital consideration. Supporting the design cycles and the analyses specific to each launch requires a reliable system. To meet schedules, CFD simulations are typically run around-the-clock every day of the year. Since the system is fully utilized, system downtime can result in a day-for-day slip that adversely impacts the other members of the design and launch team.
“We considered many things in determining the criteria for a reliable system,” said Ward. “Our previous shared-memory server lacked the growing throughput capacity required by the Delta IV program. The reliability associated with the cluster architecture was relatively unknown to us. To accommodate for this unknown, we figured into our system performance estimates the impact of potential system downtime on computing throughput.”
“We referred to this estimate as the system ‘uptime factor,’” said Ward. ”The demonstrated uptime factor of our previous system of more than 99 percent---up and running 24/7 with all 24 processors at nearly 100 percent capacity for that entire year. We didn’t anticipate that a cluster would be able to maintain a comparable uptime factor when running at maximum capacity. We developed our performance estimates for the new cluster using a conservatively low uptime factor, 60 percent, to ensure that we would end up with a system that would meet our throughput requirements.”
“That estimate proved to be very conservative,” said Ward. “With the new AMD Athlon processor-based Linux NetworX cluster system, we are seeing reliability that is far exceeding our expectations. After we became experienced and settled into a routine with the system, the reliability increased to more than 98 percent.”
“It was assumed that failures would occur more frequently due to the higher parts-count associated with the distributed-memory systems,” added Ward. “However, the failures very rarely result in a complete system malfunction. Rather, most failures result in the temporary loss of a single node, which leaves the cluster operating at close to 99 percent capacity. System performance can be quickly restored to 100 percent after a failure by replacing the failed node with a spare node that was provided by the vendor. Diagnosis of the failure and repair of the node can then be performed at our convenience without impacting the system performance. On the other hand, a hardware or operating system failure on our shared-memory system, which is admittedly very rare, results in a complete system shutdown.”
Success
The AMD Athlon processor-based Linux NetworX cluster system is being put to the test on a program that is showing great potential to become the new standard for developing and launching rockets for industry-leader Boeing.
Though the cluster system has only been in operation for about 16 months, the system is performing significantly better than expected for the Boeing Delta IV aerodynamics team. The ultimate value of the system will be demonstrated when the Delta IV rockets achieve mission success of their initial flights, and those thereafter. High-quality aerodynamics predictions will contribute greatly to the Delta IV’s successful performance in flight.
The Boeing Delta IV makes its debut in 2002 with three planned launches from the new Space Launch Complex 37 facility at Cape Canaveral Air Force Station in Florida.
The goal that the Boeing Delta IV aerodynamics team intends to achieve with the new cluster system is to increase the reliability of the Boeing Delta IV Rockets. The reliability, performance and efficiency of the AMD processor-based Linux NetworX cluster supercomputer are giving every indication that the team will do just that.
© 2002 Advanced Micro Devices, Inc. All rights reserved.
AMD, the AMD Arrow logo, AMD Athlon, and combinations thereof, are trademarks of Advanced Micro Devices, Inc. Other product names used are for identification purposes only and may be trademarks of their respective companies.
|
|
|
|
|
