Suppliers love user stories involving Formula 1 Grand Prix racing teams, particularly the winners, like Vodafone McLaren Mercedes. However, we need to separate the race track glamour from the IT reality and ask if the storage example, use case in the modern jargon, is relevant.
This SGI case study is interesting because it is a different way of managing multi-tiered storage.
It's to do with the McLaren Mercedes team and their hot new driver, Lewis Hamilton. So let's assume the hot news glamour - innovative technology, etc, etc as read and look at what is actually going on.
Jonathan Neale, McLaren Racing MD, sets the scene: “We’re competing for first place in an environment where the difference between first and tenth is about 0.6 seconds, so we’re constantly seeking fractions of a second in performance improvement. On average we’ll make a change to the car every 20 minutes during the course of a season, and to do that, simulation is vital in making efficient changes to the car.
Simulation needs data and simulation needs to run quickly so as not to slow down race car development.
In particular, McLaren is looking at the car's aerodynamics and optimising the use of wind tunnel testing through simulating wind tunnel scenarios beforehand.
Neale says: “One of the key parameters in differentiating a Formula 1 car is its aerodynamics. The car with the driver in it weighs about 600kg, and at speeds over 150mph the car generates enough downforce to be able to run it upside-down. To optimise the aerodynamics we do a lot of design work and track testing, but computational fluid dynamics (CFD) in particular has been an area of major advance for us in recent years.”
This is where the SGI storage has had an influence.
Dr Mark Taylor, Head of CFD, McLaren Racing, said: “The biggest impact that CFD has is on our wind tunnel testing programme, where the expansion of the SGI platform over the last few years has meant that the quality of components we send to the tunnel is much higher.”
The SGI kit
Like many Grand Prix racing teams, Vodafone McLaren Mercedes has been using CFD for a number of years. Fro 2005 it has used SGI for CFD supercomputing, storage and visualisation equipment. The SGI kit includes Altix supercomputers, visualisation solutions, InfiniteStorage, the InfiniteStorage shared file system CXFS, and the InfiniteStorage Data Migration Facility (DMF).
There is a need for multiple tiers of storage because of the sheer mass of data involved.
We need to go back to the scenario to illustrate this: Taylor said: "When we design parts in CFD and they go to the tunnel, if they’re positive they go to the car. But sometimes we don’t get the feedback we expect from the track. The drivers may tell us that a particular component isn’t working quite the way we expected, or could be made to work in a better way. This feedback then comes right back to CFD, where we explore it to see if we can understand and relate what the drivers are saying to what we’re seeing in CFD. When that happens, we design a new solution based on the drivers’ input."
"Frequently, if we can prove the concept works in CFD, we’ll have the part made and sent to the track straightaway, while sometimes it will go through a wind tunnel test and then a track test. This is how we’re able to maximise the impact of the aerodynamic simulations we’re carrying out in CFD on what actually happens at the track.”
The data mesh
McLaren Mercedes applies CFD to very large meshes, of the order of hundreds of millions of cells, and this creates very large data files. The issue then is how to handle all of the data. McLaren wanted a facility that would ensure that all the data for an entire year’s car programme would be available at all times, without clogging up its scratch storage facility. The Data Migration Facility (DMF) does this and was introduced in January 2007.
DMF enables cases that are older than a certain length of time to be automatically taken offline to a tape facility – making the much faster scratch disk space more or less self-managing. The system automatically takes the oldest files off first, but the CFD team are still able to access them from the tape library at any time. So, if an engineer wants to look at something from 12 months ago, they can access the file and it is automatically pulled back from tape without any input from the engineer.
In effect, as far as the users are concerned, there is a single data store which is spread across tape and disc but accessed through the usual disk data access method. If the requested data isn't on disk hen the SGI software fetches it automatically from the tape library.
Taylor said:“DMF has been a huge benefit, enabling us to look at results from throughout the year – which is very effective in terms of how our programme works. Sometimes you develop a part through CFD and the wind tunnel, but it can be some time before it appears at the track – so it’s great that we can recall old results when we want to understand what’s happening at the track or at a particular race. DMF also gives us a lot of room for expansion, so we can run larger and larger and also more transient cases, without having to worry about filling up our disks. The data is all managed back onto the tape library as required – and we can just concentrate on getting the best results we can!"
“When we had our second phase of expansion of the SGI platform, in a sense it was proving too successful, because we were generating an even larger amount of high quality results which we wanted to retain for input into the design process as we moved forward. Without DMF we were having to delete files that we would have preferred to have kept for the duration of the design programme. But when we introduced DMF we were able to retain all of the data for the entire year’s programme, which has been very effective in building up our design knowledge over the programme."
“As we develop forward, we’ll also be able to look back at the results for our MP4-20, MP4-21 and MP4-22 cars, which will be important because the rules are remaining quite stable next year, so the current MP4-22 will actually feed through a lot into the development of our next car, the MP4-23. Also looking forward, as track testing is having to be reduced, there’s even more emphasis on simulation to deliver increased performance. So we would expect that as the testing restrictions continue, the competitive advantage we receive from CFD will only increase.”
Will this work for you?
This certainly works for McLaren in terms of the turn-around speed for aerodynamic changes tits cars. One example relates to the development of a new front wing module. In less than a week, Vodafone McLaren Mercedes was able to run its first track test on the design of the wing, run the results through CFD analysis, build and test a new part in its wind tunnel, fly the new module to the test track and fit it to the car – producing a 0.2 seconds improvement in lap times.
But the average enterprise and public sector data centre doesn't need to do this. However, it will certainly have historical data and where this and current (on-disk) data need to be used together then an SGI-type, transparent, disk-and-tape tiered storage design could work well.
Users and applications just fetch data, not needing to know whether it is on disk or tape; it's just file data, in a single namespace as it were. Clearly the tape access will be slower than disk but if there is too much data to be held on disk in the first place then this is a limitation that is necessary.
In fact, with very large files, the streaming speed of tape access could compensate for the longer access time for the files, rendering the tape-disk speed difference much less apparent, perhaps nullified altogether.
Data mining might be a relevant application, but anything involving processing of data in extremely large data sets could be a candidate for SGI's InfiniteStorage and its DMF functionality.