HPC IN F1
Gemma Church takes a look at how HPC technology is assisting F1 teams at the cutting edge of the sport
In a world where one hundredth of a second can make or break a race, supercomputing has become the secret weapon of Formula 1.
Broadly speaking, there are four main areas that affect the car’s performance: the driver, the engine, the tyres, and the aerodynamic. With FIA restrictions on engine development, and teams limited to a single tyre supplier, it’s the aerodynamics where engineers can make a difference.
Improving the aerodynamics of an F1 car is no easy feat though, with engineers needing to test the vehicles in wind tunnels to check and optimise their designs, as Graeme Hackland, IT manager at the ING Renault F1 team, explains: ‘Most teams operate at least one wind tunnel for aerodynamic developments, the bigger teams two. The wind tunnels usually operate 24 hours per day, 365 days a year. Although most aerodynamic developments are performed in the wind tunnel, capacity limitations prevent expansion of the aerodynamic developments to meet performance improvements.’
Hackland adds: ‘Wind tunnels are expensive to operate, and to build a new one can cost tens of millions of pounds. Due to these limitations and advances in technology, most teams have started or have been using computational fluid dynamics (CFD) in parallel to wind tunnel testing to correlate wind tunnel data and to understand the aerodynamic flow structures in more detail.’
The ING Renault F1 team uses an Appro Xtreme-X2 system for its CFD simulations. But the ING Renault F1 team is not a one-off. Nearly every team that builds its own F1 car uses a supercomputer. BMW Sauber uses the services of Albert 3 and in the Ferrari Data Center, an Acer/IBM/Racksaver using AMD Opteron processors reduces the time of aerodynamic simulations (see panel, p27). Each team either has its own supercomputing infrastructure or uses a partner’s.
CFD has been instrumental in the cars’ changing designs throughout the 2008 season. For example, since the 2007 season it seems most teams have established that elongating the engine cover provides clear advantages in terms of rear-wing efficiency. By improving the quality of airflow directed towards the rear wing, the fin increases downforce and allows the teams to run lower rear-wing angles, enhancing the car’s top speed.
And those teams that have only just began to latch onto HPC are realising how important a technical advantage is within the world of F1. Speaking at a press conference ahead of the German Grand Prix in July 2008, Vijay Mallya from the Force India team said: ‘We have the support of EADS, who have also dedicated about 10 people to us concentrating mainly on aerodynamics and CFD where we need to play real catch-up. When they first came and assessed the team’s capabilities prior to the start of the 2008 season, they told me quite categorically that we were three years behind the game, so we are looking forward to their active support and help. We have three wind tunnels running now, one of our own and two that are leased facilities, so we are giving it all we can.’
The ING Renault F1 Team
With only 15 points from nine races, things were far from rosy for the ING Renault F1 team halfway through the 2008 season. But the tables turned and the team’s unrelenting development work began to pay off, giving them 65 points from the next nine races.
Behind the scenes, Renault’s technical department was persistent in its work, even when its competitors had already refocused attentions on the 2009 season. From the adoption of a shark-fin engine cover to dramatically revised barge boards, the R28 car was repeatedly tweaked in the latter part of the season.
This R&D effort was aided by a 38TFlop supercomputer, which was supplied by Appro to be used within CFD research at the ING Renault F1 team’s technical centre in Enstone, Oxfordshire. The solution is based on a balanced design using AMD Opteron processors and InfiniBand networking, and provides a platform to create the team’s car designs using CFD in a virtual wind tunnel.
Appro’s servers are helping the ING Renault F1 Team improve its CFD work. Photo courtesy of ING Renault F1 Team Computational Aerodynamics Research Centre.
The Appro Xtreme-X2 solution offered the team the performance and availability requirements as well as the scalability and balanced high performance and fault tolerance, according to Hackland, who adds: ‘In addition, the Appro system management environment (Appro Cluster Engine – ACE) scales to thousands of nodes and provides a new level of simple, flexible, and reliable cluster provisioning and monitoring.
‘The Xtreme-X2 series provides a balanced framework where commodity hardware can be used in a production HPC environment. The Xtreme-X2 is designed to maximise job throughput while simultaneously reducing management overheads.’
The benefits of supercomputing for the ING Renault F1 Team are quite extensive, including a 20 per cent saving on the running cost of a conventional wind tunnel, reducing the full car simulation turnaround time by a factor of three and increased CFD computational throughput by a factor of 20. Hackland says: ‘Our Appro Supercomputer, being the largest in Formula 1, offers the ING Renault F1 team the opportunity of having a competitive advantage within the field of CFD [and] increased CFD resources should improve the development rate of the car.’
Hackland adds: ‘CFD requires significantly less investment than a second or even third wind tunnel, but requires considerable computing resources. The ING Renault F1 Team is very conscious of the need for improving cost and operational efficiency, and so the solutions offered by CFD were very attractive. This was why the decision was taken to expand our CFD facilities with an investment in the new Computational Aerodynamics Research Centre. As part of this investment, a new supercomputer was installed to supply the additional computational power required to provide a virtual wind tunnel. The virtual wind tunnel will be able to deliver 50 per cent of the performance gains developed by using a conventional wind tunnel at half the cost.’
The Red Bull F1 team has always been vocal on its aim to put the fun back into F1, but behind the scenes the team has a whole host of HPC technology to help the team with the more serious business of winning races.
The team uses computer aided design (CAD), wind-tunnel analysis and complex CFD applications to develop its cars. Due to the huge demand for computing power, the organisation runs a supercomputer and uses grid computing to support the process.
Red Bull worked with Platform Computing to accelerate the design process and shorten production times in a cost-effective way. Although Red Bull Technology had completed an upgrade of its IT systems, the company found testing such CFD simulations to be very demanding on its processing power.
Red Bull Technology addressed this by entering into a technical partnership with Platform Computing and implementing Platform LSF. Steve Little, strategic account manager at Platform Computing, says: ‘Platform LSF’s software intelligently manages and accelerates workload processing for data intensive applications such as aerodynamic testing. A key benefit is the ability to identify IT resources that can be incorporated into a virtual cluster.’
Apart from supersized supercomputing centres, an ecosystem of desktop supercomputing is also appearing in F1 to help analyse and simulate wind tunnel results. ‘These resources can include servers, mainframes, and even individual desktops,’ says Little. ‘For example, when a designer leaves the office for the day, the processing power of their machine can be put to use on stress or vehicle dynamics analysis. The solution also schedules simulations and guarantees the completion of workload across a distributed and virtual IT environment.
‘Since implementing the Platform LSF environment, Red Bull Technology’s CFD department has increased the throughput of jobs by over 20 per cent and maximised the use of its existing hardware.’
Deployment of the Platform LSF environment took just five days from the time Platform Computing was on site to when the installation and configuration was completed. As part of a technical partnership, Red Bull Technology has access to ongoing technical support and can rely on Platform’s assistance any time, day or night. Before the Platform LSF environment was implemented most jobs were scheduled manually. Now all jobs can be prioritised and scheduled automatically without the need for human interaction. The result is that computer-intensive testing can be done overnight; a time when powerful processors typically lie idle.
Nathan Sykes, CFD team leader at Red Bull Technology, says: ‘We’ve been able to optimise the AMD/IBM cluster implemented last year by using the additional processing power in the network so that the team has been able to significantly speed up analysis time, achieving a 20 per cent increase in the number of tasks it can complete. Platform LSF helps us optimise designs that maximise the down-force and reduce the drag the cars create on the track. We are confident that we can continue our rise up the rankings over the next few seasons.’
The FIA is regulating aerodynamics in the 2009 season. Among the regulations, downforce will be dramatically reduced and the cars’ bodywork will appear much cleaner, due to new regulations that effectively outlaw extraneous items such as barge boards, winglets, turning vanes and chimneys.
As well as reducing overall aero performance, the revisions are also designed to increase overtaking by making the car less susceptible to turbulence when closely following another driver.
Regulations aside, FIA president Max Mosley also recently asked the Formula 1 teams for proposals on how to cut the costs of competing in the sport by 50 per cent. According to a statement from the FIA: ‘Among possible cost savings which the teams may wish to consider are: restrictions on simulators, wind tunnel use, CFD and other home-base facilities.’
But how will such cost saving plans affect the CFD simulations within F1? ING Renault F1 Team’s Hackland remains optimistic of the team’s future need for supercomputing, and says: ‘The ING Renault F1 team’s stated aim is to increase its competitiveness and produce a car that is capable of challenging for the world championship next year.’
Hackland adds: ‘We see computation growing within the ING Renault F1 team to other areas of the company – including design, race engineering and vehicle technology groups. In addition, the aerodynamics group’s requirements will continue to grow in the future.’
Ferrari turns to Windows HPC
By Tom Wilkie
Microsoft recently underlined its determination to be a significant player in all applications of high-performance computing by choosing the high-tech setting of Ferrari’s Formula 1 test circuit in Maranello, Italy, for the European launch of its Windows HPC Server 2008.
Aerodynamics of the car body is currently the biggest application of HPC in Formula 1 car design and development, according to Piergiorgio Grossi, IT manager for the Ferrari racing team. The process involves not only design, but testing of prototypes in a wind tunnel, processing and analysing the data and corroborating these results with tests of the cars as they behave on the track.
Ferrari makes extensive use of the Fluent CFD package from Ansys to carry out its aerodynamics computer simulations, and Pelle Olsson, HPC product marketing manager for Microsoft, illustrated how the server system can make it possible to run the program on clusters with no more effort than was needed for serial versions. By clicking a few options on the Fluent launch menu – notably enabling the Microsoft Job Scheduler – the management of the cluster is carried on in the background, making it possible for engineers to concentrate on devising the program and not optimising the HPC cluster configuration. ‘The point is that you haven’t seen Microsoft software – all you’ve seen is Fluent,’ Olsson said.
Grossi said that ease of transition was one of the reasons that Ferrari had decided to opt for Windows Server 2008: ‘It’s easier for us to manage it. It’s compatible with our ecosystem and our engineers are used to it.’ He pointed out that the Formula 1 business was so competitive that cars have a lifetime from design to their last race of only 18 months.
Moreover, the car evolves within that period, both as the result of general improvements and because packages are tailored to the characteristics of each individual race track. Computing power is crucial in giving the Ferrari team its competitive edge, he said. ‘Our use of HPC is peculiar – we run the servers 24/7. We want to have total availability.’
Grossi said that parallelising legacy software to get the best of our modern HPC was a problem for the company. Ferrari had several tens of applications that it had developed itself, in-house. In particular, many of them involved the telemetry of transmitting data back from test cars and also track-specific ‘decision support software to help engineers decide on strategy and modifications to the car appropriate to the different characteristics of different race tracks. In addition to Fluent, Ferrari’s engineers use Matlab from The MathWorks and other specialist software but ‘we have also built a lot of code using Excel and would like to spread it over more processors,’ Grossi said.
Ferrari is continuing with its Linux-based server systems as well, and clearly will not be moving away entirely from the ‘handcrafted’ approach to high performance computing.