Formula 1: The super-fast net driving teams to the podium
bbc.co.uk
bbc.co.uk
Data coming from the vehicle sensors and other sources are concurrently being used for analysis (by both hundreds of engineers and a wide range of "algorithms") as well as input to simulation models. The results from the simulations means that even more data is being generated even when the vehicle isn't running.
Within minutes, using the data, the different engineering groups (typically responsible for a sub-system, ie. engine, tires, aerodynamics) arrive at conclusions which then the vehicle's race/performance engineers are using to enhance the setup of the vehicle. The results of changes are then fed back to the engineers, and evaluating if the analysis and predictions are correct is a big part of the post-event work.
I honestly can't think of any other industry that carries out this kind of analysis of highly non-linear systems at this scale and speed. The only other industry that I can think of is finance?
Disclosure: I work in motorsports
I assume this is in reference to use of the term "race engineer" in F1.
I can assure you that the sport does use real engineers, and very intensely so. Folks like Adrian Newey would qualify as engineers by anyone's definition, and each team has heaps of MechE's, EEs, software engineers, even materials engineers, back at home base.
They make the tactical and strategic decisions on when to pit during the race as well as guide the car set up and tuning on Friday and Saturday practice.
Here's a few impressive videos from McLaren's mission control, located in their base in Woking, near London.
https://www.youtube.com/watch?v=vYhl7csZJHw https://www.youtube.com/watch?v=Ple7W6bdxJc
Well, yes. From the article: "the FIA - restricts them to 60 staff members at the trackside, including engineers"
Does that include the people at the factory in the mission control? I don't think it does. McLaren's Mission Control (in the videos) already has about 20 people working.
I didn't even count the actual pit crew in my estimate of 50. There's around 20 people in the pit crew.
Right, that was the point of the article... they can only have 60 people trackside, so they need all this bandwidth to send the data back to the factory where even more people can analyze it.
This is absolutely terrible. Your network should be secure, even if you tell everyone how it's done, and reluctance to do so makes it very likely it's not secure.
Here's Steve Bellovin's thoughts on this:
'The subject of security through obscurity comes up frequently. I think a lot of the debate happens because people misunderstand the issue. It helps, I think, to go back to Kerckhoffs's second principle, translated as "The system must not require secrecy and can be stolen by the enemy without causing trouble," Kerckhoff said neither "publish everything" nor "keep everything secret"; rather, he said that the system should still be secure even if the enemy has a copy.
In other words – design your system assuming that your opponents know it in detail. (A former official at NSA's National Computer Security Center told me that the standard assumption there was that serial number 1 of any new device was delivered to the Kremlin.) After that, though, there's nothing wrong with trying to keep it secret – it's another hurdle factor the enemy has to overcome. (One obstacle the British ran into when attacking the German Enigma system was simple: they didn't know the unkeyed mapping between keyboard keys and the input to the rotor array.) But – don't rely on secrecy.'
On the other hand, if it is my information you are securing and if I don't have good reasons to trust you, I want to know how it is being done, even if that means attackers also know. If everyone is your client (for example, if you provide public services), then everyone must know, so they can independently audit the system.
Obscurity can be a layer of security in one system, made ,managed and audited by trusted entities, but, generally speaking, it is a weak layer for a attacker with a great enough motivation.
On the other hand, obscurity is detrimental to a collective of systems made, managed and audited (or not) by a great variety of entities. Sure, on the real world we place trust on companies to handle their security well, but that has ended badly in the past. The knee jerk reaction to security through obscurity we have is beneficial because it is a symptom of security issues in the system.
In conclusion, regarding obscurity, the beneficial effect of an extra security layer is only greater than the potential malefic effect of hiding security problems if the entities behind it, including the auditor, are competent and trustworthy. As a rule of thumb it should be avoided or have its bad side mitigated by independent security audits.
That's not how things work
Remember: disagreement should not lead to downvotes, it should lead to a reasoned reply. The reason for a downvote is when a comment detracts from the discussion, and I can't imagine how the parent post could be viewed that way.
But this is F1, everything is a secret. F1 teams go to crazy lengths to keep anything that could be an advantage to themselves (ex: Mercedes mass-damper, Red Bull double-diffuser). I would suspect that there is a very strict NDA as part of supplying the communications to each team.
Interviewee: "sigh. There are multiple layers of blah-blah, with foobar authentication schemes. That's all the detail I'm prepared to go into."
<Reporter writes down, "tight-lipped about security">
AT&T and Tata don't owe you an explanation just because you think that's the way the world should work. Such an explanation takes time away from doing other things, all so only a tiny fraction of a percentage of the world population can actually understand what was said.
there's a difference between relying on obscurity, and publishing all of your designs of a private system.
the underlying design principle being promoted is to assume that your design specification has been compromised - not to compromise it yourself.
Once a year they would come around and give a talk about asset reliability and how they operate.
An interesting story was how the one time they ignored their data (because, due to the reliability of the part involved, they couldn't believe the likelihood of such an event occurring) they had a catastrophic engine failure mid-race.
Thanks to the data collected they were able to run detailed simulations to "debug" the issue - a small change somewhere else causing a part to vibrate at it's natural frequency - i.e shaking itself to bits. This then helped them mitigate the risk of the same thing happening in the future.
From then on they always trust the data.
Test your sensors, trust your sensors, trust your data.
It’s really the engineers making the pre-race setup for the car and making real time adjustments during the race that makes a winning F1 team. It has been argued the F1 drivers merely are “along for the ride” and it’s really what goes on behind the scenes that wins a race. https://www.youtube.com/watch?v=1j9r7Ue6XnA
"8.5.2 Pit to car telemetry is prohibited." source: http://www.formula1.com/inside_f1/rules_and_regulations/tech...
"F1 drivers merely are “along for the ride”" This is wrong on so many levels. F1 is a very physically demanding sport. Their reflexes have to be as good in lap 40 as they were in lap 4. They need to perform changes to the car in real time, mostly between turns, as they're told. They need to know how to attack, defend, overtake, save tyres, save fuel.
The difference is that 10/15/20 years ago they did all this without not much data, and now they have much more context to each decision.
You clearly do not watch much Formula 1.
So that's 800 gigabits per second, nearly a Tbps. Which is a lot of network throughput. Although this is not impossible, it seems surprisingly high, and I wonder if the article should have said "100 gigabits each second", which is still a lot of bandwidth but a little more reasonable.
The article describes the data being transmitted as largely telemetry. Let's say they are sampling each of their metrics every 10 milliseconds (100/second), and they need a 32 bit integer to store each value. So, that's 100 * 32 / 8 = 400 bytes per second per metric. Divided into 100 gigabytes, that would be 250 million metrics. Even if you change these assumptions, clearly this link is used for more than telemetry.
800 Gbps requires a large number of computers to generate and transmit, then another large set of machines at HQ to ingest all that data and then run analytical models against it. Just to handle the network traffic, without any data analysis, if you assume servers have a pair of 10 Gbps network interfaces, that's 40 machines.
If you assume hard drives can write at about 100 megabytes per second, that's 1,000 disks just to handle the write IOPS. But, with those 40 machines, that's an average of 25 disks per machine, so not unreasonable. With SSD's it would be fewer devices.
I do not know if a way to get this much network throughput over a radio link, so I can't imagine it's all coming from the car itself.
It would be interested to know what other data is transmitted over this link during a race other than telemetry.
The link will be utilised for telemetry, data services, radio intercom, video and other connectivity.
There's also high-definition imagery from each car, I guess that some of it is generated in the car and the other imagery is supplied by the TV cameras.
I don't know though if these run on UHD, but given the boatloads of cash behind F1 I'd assume so.
What am I missing?
The article mentions it also needs to span continents. I'm pretty sure that if you want to reliably transfer 100GB/s, you'd need to roll new cables between a couple of continents. North-America to Europe might have the capacity, but I'm sure (looked into it before[3]) that from South America to Europe can't do it directly even if you exclusively rented the cable that runs between the continents (which is the Atlantis-2). So you'd need to route through North America or maybe Africa, which adds latency. Probably still under 300ms single-trip, but it still seems unfeasibly large to me.
Even if the article is mistaken and means 100 gigabits instead of -bytes, I have a hard time believing it. It's possible, but hard to get that much data between continents while keeping the latency low enough, at least without rolling your own underseas cables which even Google doesn't do on their own[2].
[1] https://ams-ix.net/technical/statistics
[2] http://gizmodo.com/google-funds-an-undersea-cable-to-connect...
"The Tata Global Network(TGN) has Trans-Atlantic and Trans-Pacific data transfer capacity of one terabit per second." http://www.formula1.com/news/headlines/2012/2/13043.html (Two years old)
Pretty sure the actual number is a lot lower than that. I'd say its probably more like a low latency 1-10Gbit/s connection per team.
This may give the wrong impression, so just to be clear there are strict regulations in F1 regarding the reuse of components and static designs.
http://www.formula1.com/inside_f1/rules_and_regulations/spor...
Mercedes, who sit in #1 and #2 in the driver standings, and of course #1 in the constructors standings, is presumed to have an advantage because of the way they laid out their turbochargers.
https://www.youtube.com/watch?v=NuBB2F6IutQ
By F1 rules, no other team can change their own engine designs until next season.
Next week at Spa (a very fast track) we will see radically different cars from, say, Monte Carlo (the slowest track). The cars look visually different if you look at them in side by side images.
Teams come out with a season design, and from that they adjust wings, tire pressure, gear ratios or change points, etc, within the tolerances allowed. I've only marginally been following F1 for the past year, but I'm wholly unaware of any substantial construction change from track to track.
Edit: To substantiate this a bit, here's the Ferrari car at Hockenheim (a very high speed track), compared to Monaco (a twisty, "urban" track).
http://www.zimbio.com/pictures/bGZpJEcephz/F1+Grand+Prix+of+...
http://www.formula1.com/news/headlines/2014/5/15851.html (click on the Ferrari picture)
This year the set of gears is fixed over the entire year (there's some wiggle room the details of which I forget, but that's the gist). A team has to elect at the beginning of the season what gearings they'll put into the gearbox.
The tradeoff for this was that this year they've got 8-speed gearboxes compared to last year's 7. So in the past they had to pick 7 gears optimal for one track, now they've got 8 gears that must compromise across the whole season.
whoopee doo.