Development at Audi Shifts Into a Higher Gear
An expanded blade server cluster from HP offers greater computing power for crash simulation tests at Audi. The new system delivers peak computing speed of 29.184 Tflop/sec, thereby greatly speeding simulation process and admitting it to the elite club of the 500 fastest computers in the world.
German cars are among the safest in the world. To make sure that they remain this way, engineers in the development laboratories of automobile manufacturers simulate traffic accidents of all kinds. The angle of impact and the design of energy-absorbing structures, the rigidity of the passenger cage and the breaking strength of the frame all pay a decisive role during impact. Experienced design engineers employ computers to vary the technical parameters in order to study all aspects of a collision. Audi conducts 5,000 simulations per week. Each model undergoes a 48-month development phase involving 1,000 simulations. The objective is to use the acquired information in building the safest possible vehicle.
Simulation programs help technicians analyze and optimise crash behaviour. What’s more, simulation tests offer automobile manufacturers both technical and financial benefits. Because simulation applications make testing with real automobiles superfluous and eliminate the cost of developing expensive prototypes. The result is that development times are getting continually shorter, whilst the range of Audi models grows.
Crunching numbers instead of bending fenders
Nearly all simulation calculations involve the processing of tremendous volumes of data by programs called “solvers.” And this is where computer clusters shine. The many blades in a cluster deliver the computing power and speed needed to crunch the numbers in the shortest possible time. Because no development engineer wants to wait days or even weeks for the results of his crash simulation tests. And no German automaker can afford to launch a new car model several months late on the highly competitive global market with its short development times. Those who are late risk being punished by customers and a loss in revenues.
Blades win out
When the leasing contract for development at Audi expired after three years, the company took the opportunity to significantly boost computing power to meet the increased demand. Audi began using a blade cluster from HP in August 2007 and was quite satisfied with the service and the speed in getting the system up and running. HP was thus commissioned to replace the existing cluster from another manufacturer. An additional 288 model BL 460c servers and Voltaire Grid Director™ 9024 20 Gb/s InfiniBand switches were added to the existing cluster of 320 servers. Voltaire Grid Director switches use 20 Gbps InfiniBand technology, which provides high bandwidth and very low latency to enhance the infrastructure and application performance. The switches provide the throughput and scalability needed to capture the full power of the servers to calculate the tremendous volumes of data generated during the simulation process. The existing server cluster and 240 of the new servers form a platform for operating PAM-CRASH simulation software. Forty-eight blades run primarily STAR-CD, an application for fluid dynamics simulation. The new HP cluster now delivers 29.184 Tflop/s of peak computing power, thereby vaulting it to no. 81 in the current top 500 list of the world’s most powerful computers. Intel Dual Core processors and high-performance software applications deliver greater computing speed. Dual core technology employs two processors in a single processor socket, each with its own independent registers. The advantage of dual-core processors is that they combine two complete execution cores in one physical chip. Each core has its own arithmetic logic unit (ALU) and register and can perform calculations on a stream of data independently of the other core. This makes it possible for the program to process two sets of data at the same time, thereby significantly increasing calculation speed, especially when coupled with high-performance software.
HP as prime contractor
There was another reason for entrusting HP with system expansion. Audi IT managers wanted to optimise their infrastructure and reduce the total number of suppliers. The benefits of this strategy are illustrated by this project. The capacities required for PAM-CRASH and STAR-CD could be assigned within the server cluster itself, because both applications share the same platform. This is especially advantageous when it comes to PAM-CRASH simulations. STAR-CD is used less frequently, thereby making more computer power available for other tasks running in this system. Audi selected HP as the prime contractor. The automaker was thus able to profit from the experience HP had gained in installing and commissioning the first cluster. The result was that the project was completed in much shorter time. HP conducted benchmark tests at its Benchmark Centre in Böblingen between Christmas 2007 and New Year’s Day 2008 aimed at thoroughly testing the cluster servers and software. The installation of the cluster was begun in January. HP delivered the fully configured cluster to Audi in April. The cluster in turn was installed in only a short three weeks. The system at Audi’s CAE laboratory (Computer Aided Engineering) was put into operation in May.
Lower energy consumption
Blade technology also reduces energy consumption. The system thus generates less heat and requires less cooling. The blade system installed at Audi consumes only 86.7 kilowatt hours of power whereas a comparable rack system would have required 115.57 kilowatts – or 33% more energy. The reason is to be found in the special construction of HP c-Class BladeSystem architecture, in which Dynamic Smart Cooling technology is employed to manage energy consumption and regulate cooling. Thermal logic technology dynamically adjust energy consumption and cooling, thereby leading to lower energy consumption overall. The construction of the blades offers an additional advantage. The compact blades require much less space than an equally powerful large-rack system. At Audi the blade cluster requires only seven enclosures whereas the traditional option would have required nine racks. In addition, using Voltaire InfiniBand for the interconnect helps keep energy consumption to a minimum. Even with 20 Gb/s performance, Voltaire switches require on average only five watts of power per port, making them an attractive technology for energy efficient solutions.
Plug & play technology also simplifies installation. The blades are simply inserted into slots in the blade enclosure where they plug into the backplane, which contains all the required cabling. Plug & play technology takes care of the rest. At Audi the blade solution reduced the cable requirement by 50%. Another advantage of the blade solution is its outstanding scalability. Blade servers can be easily expanded to meet growing requirements. The administrator needs only to insert additional blades into unpopulated slots to generate added computing power. Audi development engineers are currently testing the crash characteristics of their automobiles on a total of 608 BL460c blade servers running RedHat Enterprise Linux. This gives Audi a decisive advantage on the rapidly changing world automobile market.
