Country missing? Please select your nearest region...
Airbus is taking industrial simulation to the next level with its supercomputing operations designed for building the A350. Penny Jones visits Toulouse to find out how much of a game changer such high-end compute power can really be
15 June 2012 by Penny Jones - DatacenterDynamics
The city of Toulouse, with its pink buildings wrapping around the Garonne river, has a culture steeped in history. It is known as the unofficial capital of the Occitan (the historical region of southern Europe) but Toulouse could never be accused of living in the past. The city is full of universities - Toulouse University being the most famous – and all it takes is dip into the outskirts of town to realize you are entering an environment dominated by students pedalling push bikes.
Toulouse has another title. It is also the center of the European aerospace industry. Both accolades feed off each other in a cycle. The Galileo positioning system, the SPOT Satellite system, CNES’s Toulouse Space Center, Thales, Astrium and EADS all have headquarters in Toulouse. EADS’ commercial aircraft subsidiary Airbus Industrie also has its headquarters there as well as its final assembly line for the A300, A310, A320, A330, A340 commercial aircraft and the giant flagship double storey jet liner, the A380. Like so many other companies, its key reason for setting up such critical establishments in the city is the availability of skills on the ground. Many of these companies are moving so far ahead with the way they use technology in aerospace design that it would be difficult to be positioned away from such a resource.
At Airbus, the A380 with its size may have the headlines but its newest design – the A350 – is the one breaking barriers for teams at Toulouse. From streamlined design to IT and assembly, Airbus is entering new territory with production on its latest long-haul jet. In the process, it is deconstructing barriers in the area of simulation with a speed that would never have been possible without innovations in high-performance computing (HPC).
Towards a virtual aircraft
In recent years advancements in technology and reduction of costs have made simulation a differentiator in the aerospace market. Deals are often made on an aircraft’s time to market and the ability of the end user to customize their purchase. Fuel efficiency is also key.
HPC has gone from being used for high-speed computational fluid design (CFD)-based scaling and simplified geometry to the complete physical modelling being used to create the A350 jet liner. A new-generation jet being designed to be 25% more fuel efficient, the A350 uses advanced materials and has new Carbon Fibre Reinforced Plastic fuselage for easier maintenance.
The power behind Airbus’ simulation is provided by HP 40c Performance Optimized Data Centers (PODs) run as part of an outsourcing deal at Airbus’ headquarters in Toulouse and facility in Hamburg. Airbus previously ran supercomputing out of a hosted environment, because supporting such capabilities in house was too expensive and challenging. With a containerized solution, however, HP has been able to deliver what is now listed as the 29th largest computer in the world, and the largest industrial HPC environment in existence today.
Despite the PODs sitting across two sites (three 40ft/12m PODs are in Toulouse and another is in Hamburg) the Supercomputing TOP500 list treats both as one facility, as they are designed to work in unison.
The system is designed for number crunching, to deliver a real-time simulated view of every part of the aircraft design to Airbus engineers, customers and parts suppliers which work from offices around the world. This is delivered through the ‘Airbus cloud’, which offers a virtual prototype that changes along with design tweaks. Before a sheet of metal is even delivered the A350 will be produced as a single digital mock-up (DMU) (or computer-generated model) which acts as a master reference throughout the aircraft’s lifecycle.
Airbus says by using the DMU, it has been able to speed up the development process, allowing information to be shared in real time and simulating the manufacturing process. It also helps customers train maintenance teams before they have even received an aircraft.
Airbus used this same process for some components of the A380 but extended it to the entire development cycle of the A350. For future aircraft, it says it hopes it will soon have the ability to create what it calls the ‘virtual aircraft’, which can be tested in real-time simulation offering a virtual certification before physical production has even begun.
Airbus SVP of Flight Physics Alex Flaig says HPC has now become critical to the design process. “We have made HPC central to our design to overcome time pressures and to get more design detail into the development of the aircraft. It has been so successful we are now looking at putting all our design software in an HPC environment in future,” Flaig says.
Supercomputing in a POD
Toulouse residents are familiar with large hauls blocking roads on their route in to town. Sections of the A380 were built in Germany, Spain and the UK then transported to Toulouse for final assembly using modified road networks catering for giant convoys carrying parts in preassembled pieces.
HP has a similar approach to design. It constructed its PODs abroad, at its Erskine site in Scotland (HP now manufactures its PODs in Kutna Hora, Prague). Once constructed, these were shipped to France and delivered by truck, with HP adding the final touches on site. The racks inside the POD were removed for transportation and installed onsite by HP. The first POD was delivered in July 2009 and a second to Hamburg in 2010 and last year Toulouse received its final two PODs.
HP took four months to deliver the first POD for Airbus. It won the contract as part of a four-year deal, arranged to allow Airbus to make the most of new hardware upgrades.
The plan is to roll out new equipment based on Intel and Nvidia’s product roadmaps, which Airbus Head of IT Projects Marc Morere says can have a large effect on the capabilities of simulation teams, and energy efficiency. “We have big dependencies on the software and architecture behind it, and new products can have a great effect on how we work,” Morere said. “Intel works on new projects for HPC every 18 months, and if you want improvements and new technology, you have to follow their tests.”
An example of how scalable this arrangement can be lies in the updates that have taken place to date. In 2009 Airbus used one cluster of 100Teraflops Nehalem servers, in 2010 it added another cluster and in 2011 a new cluster of 300 Teraflop Westmere.
Despite requiring a lot of high-end compute, one POD at the Toulouse site has still not been turned on and the PODs that are running still have room for more servers. Each POD can cater to 22 52U racks. At present they contain a cluster of HP ProLiant BL280 G6 blade servers.
Each user accessing the grid has associated licence and access agreements, and security is managed by Airbus through an agreement with Astrium.
HP manages the power at the site through its own contract with EDF at Toulouse – at Hamburg this contract is managed by Airbus.
The HP POD is directly connected to the Airbus LAN to avoid network latency risks and ensure the supercomputing operations adhere to Airbus’ security requirements.
The benefit of outsourcing
According to Airbus VP of Infrastructure Jean-Luc Vincent-Franc, the project is a “best-in-class” example of high-density computing. “With this project we have a very short lifecycle in terms of technology, and we are trying to leverage that to get new versions of technology every time it is possible. We are also much greener – this data center, compared to Airbus’ other company facility, is much more efficient (HPC3 has a PUE of 1.25),” Vincent-Franc says.
HP EMEA VP of Industry Standard Servers Florian Reithmeier says the modular facility’s ability to bring compute to the customer has been important for Airbus’ cloud model. “It is about the way people consume services, and being so close to the Airbus data center that holds all the information systems helps that,” Reithmeier says.
The project has been so successful, according to Morere, that HP would like to use HPC PODs at its other design center in Bangalore, India – but power provision is still a hindrance. “The demand for the services, now engineers can see how they work, is very high. Demand goes up with every machine installed,” Morere says.
The competitive edge
Airlines are struggling with rulings against carbon emissions, taxes, and the global recession and delivery of a new energy efficient aircraft that promises better aerodynamics and less fuel consumption can have a strong effect on bottom line. Airbus had delayed the launch of the A350 from 2014, and while it says it is still breaking barriers on its time to production with this project, the loss of orders highlights just how important time to market is in such a cut-throat industry. Airbus is hoping to deliver the A350 about 70% faster than it has any other aircraft.
Airbus revolutionized the way it did product lifecycle management as part of the A350 project. Instead of building much of it inhouse, large packages of work were outsourced to tier 1 suppliers, which makes Airbus’ product lifecycle management (PLM) systems critical to the success of the project, according to Airbus head of configuration management for the A350 Pierre Panfili.
“This means we have to always be vigilant in terms of the sizing of the IT infrastructure and be sure the software is always capable of delivering the service. We compute 1,500 mockups (of the A350) per section per day – this is the level of exchange that takes place across our teams. Next year, as we get close to launching the structure of the A350, this will be more than 8,000.
Airbus is hoping its new PLM environment will help it drill down further into the design of the A350m, according to Airbus’ Flaig. “With modelling like this we can measure the lift forces, direct force and the global force. When you have the CFD calculation you can recognize more suction, pressure, and you can get a much more detailed analysis of the features and configurations and discover where you can make more changes in detail,” Flaig says.
“The DMU is not only a representation of the physical aircraft, but contains all of the data surrounding the parts and structure though production. It even offers a 3D view to show maintenance staff how to access parts and provides manufacturers and engineers with a common reference.”
HPC3, as this project is called, is only the beginning for Airbus. HPC4 is already on the cards, according to HP Global EADS Sales Director Jean-Paul Bonnet. While not assured of the contract. HP is launching its RFP, which this time will use its EcoPOD design, which HP claims is the world’s most efficient data center. Built in Kutna Hora and transported on truck, the EcoPOD will be double the size of the PODs already on site at Airbus and use free cooling, with a PUE of 1.07. The modular solution, however, will take more design time onsite.
HP plans to include its HPC intelligence in the new RFP, as well as innovations in the network. “The challenge for HPC is to bring the electricity to the compute center,” Bonnet says. “HPC4 is also targeting a Petaflop for five years’ time.”
In the meantime, Airbus and HP are ensuring they will have staff on the ground to cater for the shift in computational power and its more widespread use across the company. Airbus is constantly engaging the education sector in Toulouse, which has had access to the supercomputing facility, just one a number of high-end clusters in the region.
This article first appeared in FOCUS 22 – out now. To sign up for our digital editions click here