Next time you see a sporty Audi A8 cruising down the road, you'll know that Linux played a role in designing this popular car.

Throughout its storied history, Audi has taken on the role of automotive technological pioneer. So when the German car manufacturer started designing the new Audi Space Frame, a high-strength aluminum frame structure designed for greater safety, increased performance, improved handling, and lower fuel consumption, this product was no exception. The only mass-produced car frame made completely of aluminum, the Audi Space Frame is enjoying much success on new models of the A8 and A2. However, Audi wanted to further enhance the safety of their cars and recently embarked on designing the fourth generation Audi Space Frame – an encounter that led to the company's adoption of Linux.

The Design Process
How a car part comes to fruition is a long, arduous task, involving many different disciplines like design, crash tests, and metallurgy. The parts for the Audi Space Frame are made from pouring liquid aluminum into a mold, which is then solidified into a casted car component. This seemingly simple task can be problematic as minor differences can dictate the product's failure or success. The mold design determines how the liquid aluminum fills the cavity and subsequently solidifies. If a particular design shows areas of the component with highly different solidification times, weak spots can occur that jeopardize the safety of the part.

The traditional way to design and build a mold is to produce several casting simulations to roughly confirm the chosen concept, build a prototype, and run tests. When desired results aren't achieved, a new design is created to try and fix the problems. This approach was successful in the past; however, it was very time consuming and the mold's success wasn't confirmed until the design had already been finalized. To incorporate any changes into the design after this point was costly and time consuming. More recently, the efficiency of this approach has been increased by the extensive use of simulation techniques, such as computational fluid dynamics (CFD), in the early design phase of a car.

The Importance of Simulation
CFD is a broad field encompassing mathematics, physics, chemistry, and computational methods. In a computer-aided engineering framework, CFD is used to simulate the motion of fluid within or around solid objects. The main principle behind CFD is that the region of flow is turned into discrete volumes, generally cubic-like cells. The governing fluid equations are then used to set up a large matrix, which can then be solved using a computer system.

A specific niche within CFD is casting simulation, where applications, such as MAGMASOFT, can create simulations of different mold designs and retrieve information from the filling and solidification process, as well as mechanical properties, thermal stresses and distortions. Casting simulation applications provide complete information on the casting process and make it easy to detect problems and correct them. Each simulation provides far more information than physical testing, making it possible to visualize the flow conditions throughout the mold, and zero in on the issues that are preventing the design from creating a perfect component.

"Casting simulations not only affect production aspects, but also play an important role in crash simulations," said Erich Blümcke, casting simulation specialist at Audi. "We are making strong efforts to increase the accuracy of crash simulations by taking into account a non-homogeneous distribution of mechanical properties of the casted part."

Audi had used casting simulation applications in the past; however, to produce one simulation of a cast on their existing Unix-based computing system would take two weeks, an unacceptable and costly time frame. This restriction limited the value of the simulation technology. To successfully and efficiently design components for the new Audi Space Frame, Audi needed to deploy a casting simulation application on a high-performance computing system.

Investigating Linux Clusters
When Audi started investigating new computing architectures, the highest priority was given to the reliability of the system, which had to be proven by benchmarking procedures and documentation of other successful installations. The second priority for Audi was the price/performance ratio. Audi needed a computing solution that was powerful, yet fit within their budget.

Linux Networx, a provider of Linux cluster systems, had been working closely with MAGMA, the producers of MAGMASOFT casting simulation software, to produce a cluster optimized for the CFD market. The Evolocity MAGMASOFT cluster system is a comprehensive system designed specifically to improve the speed and accuracy of the casting process. Audi was drawn to the Evolocity MAGMASOFT solution as it was designed specifically for casting applications and offered a compelling price/performance ratio.

MAGMA was also using Linux Networx cluster technology in-house for its own consulting engineering service. With the installation of a 32-processor Evolocity cluster, MAGMA was able to reduce its response times on complex projects and lend its computing power to MAGMASOFT customers.

Rigorous benchmarking tests then ensued on a Linux Networx Evolocity system deployed with MAGMASOFT at Linux Networx GmbH, the European division of Linux Networx. The Evolocity system ran one of the largest MAGMASOFT data sets ever, processing 200 million cells. An average number of cells typically run on a cluster ranges from 100,000 to a million. Because Linux Networx was the only vendor able to meet Audi's specifications and produce successful benchmark results, Audi purchased a 32-processor Evolocity cluster. Audi's cluster included 32 Intel Xeon processors and a Myrinet 2000 interconnect.

"Linux Networx exceeded our expectations by not only providing an extremely reliable system, but providing a turnkey cluster system preloaded with MAGMASOFT," said Blümcke. "The fact that MAGMA decided to develop its future software on a Linux Networx cluster demonstrates the capabilities of this cluster in the engineering design market."

Linux Clusters and CFD Simulations
Companies involved in engineering design and CFD use Linux clusters for their power, flexibility, and scalability. Today's rapidly changing automotive industries are shifting towards open source software platforms using commercial off-the-shelf (COTS) computer components. By using Linux as the operating system and hardware based on standard x86 architectures, Linux clustering is the culmination of both of these concepts. Linux clustering leverages the power of Linux while harnessing the power of low-cost COTS components to deliver a solution that is powerful and reliable, and delivers an improved price/performance ratio.

Reliability and scalability are other features associated with Linux clusters that make this computing platform attractive for casting applications. Scalability allows Audi to start with a small cluster and seamlessly add nodes as processing demands increase. The risk of losing an entire processing run can be eliminated when the cluster is designed to be forgiving of node failures, and a portion of computational capacity may be lost without compromising the data or completion of the task.

Linux clusters are also popular with organizations involved in engineering design because of their parallel architecture. Cast simulations are divided into different sections and each section is assigned to an individual node, with all nodes having their own memory. Linux clusters are able to coordinate the different processes and establish fast communication between the nodes. Simulating the filling process has the added complication that the computational volume filled is constantly changing as the casting cavity is filled. Therefore the load on the various CPUs has to be reassigned continuously. If this is not done, some of the nodes would be idle most of the time and the potential benefit lost. The Evolocity MAGMASOFT cluster automatically queues the jobs, distributes them on the number of nodes selected and assembles the results to make them available to the front-end workstation for evaluation.

Managing the Cluster
With a 32-processor system, Audi wanted to ensure that each node was operating at maximum efficiency, but didn't want to spend time and resources having administrators monitoring the cluster for optimal performance. Audi was able to easily combat this problem by installing a complete suite of cluster management tools from Linux Networx. Clusterworx, a comprehensive cluster management software solution, and ICE Box, a cluster management hardware appliance that fully integrates with Clusterworx increases system uptime and tracks cluster performance.

"There was a situation during the summer when the HVAC unit of our server room failed on a Saturday, so no one was there to correct the problem. When we returned on Monday, the Linux Networx cluster was the only system running fine, all the other servers were shut down due to cooling problems," said Blümcke. "The cluster management tools kept the system healthy in this situation, even though no one was there."

The Results Since installing the cluster at their site, Audi has seen significant performance gains that were impossible to achieve with their existing machine. Casting simulations that used to take two weeks now finish in two days.

"A rapid simulating time was a prerequisite to reduce the total turnaround time of even large problems to two to five days in order to integrate the casting simulations into the general product development chain of new cars," said Blümcke. "The Linux cluster enables Audi for the first time to accurately simulate the filling and solidification process of large structural parts."

Because of Audi's move to a Linux Networx system, Audi estimates that the faster simulation times, as well as the higher accuracy of the simulation results, have the potential to save money equivalent to the costs of one mold and also decrease the number of rejections, thereby reducing the costs of the car.

"We are getting involved in the design of about two to three structural parts that are relevant for crash worthiness. The handling of these parts was simply not possible on the Unix hardware platform," said Blümcke. "Only the use of the Linux cluster enabled Audi to tackle the integration of casting simulation into the design process of new cars."