Steve wonders why you need two days of supercomputing power to work out the effect of a peloton, and attempts to make older Macs work better.
Steve wonders why you need two days of supercomputing power to work out the effect of a peloton, and attempts to make older Macs work better
If there’s one consistent trend in business networking these days, it’s that everything is bigger. From the vantage point of the long, hot summer of 2018, it’s easy to see the reason. The Internet of Things has come of age in the past couple of years, making all the preceding efforts at computing and storage seem puny, shrunken, introverted little projects. IoT is a monster, no matter what your business or your intended purpose.
Take the guy I met who made wire for windings in electric motors: billions of kilometres of the stuff. His IoT project was all about industrial sensors, of which he had an average of 20 on each wire-making machine, of six distinct types. Just for a bit of wire!
Since then, my rule of thumb has been that initial estimates of the size and nastiness of a problem, when the world of computing has to suddenly handle data from the world of everything else, will definitely be wrong. Even seasoned IT types should tread carefully, because the old rules about data provided by humans tend to be limited by input speeds (typing, and so on) and get a free ride from pre-processing. By that I mean, “press a button when you see a green car” is an easy instruction to give a human – but turns out to be hugely difficult when the button-presser is a machine.
Taking people through the things that are possible in computation, and the things that lay forever out of reach, remains a concern. The simplest example is network traffic analysis. Everything that traverses a network comes from a computer of some kind. So it’s a known format and presentation, and it isn’t an analogue value – it’s a digital file-dump of predefined bits, like a pile of LEGO on the bedroom floor. Nonetheless, network traffic analysis is like taking a cool drink from a blasting fire-hose.
Or consider this non-tech example. Walmart gives its customers discounts, using vouchers printed in magazines. It turns out to be utterly impossible to figure out which magazines you should buy, and when, to maximise the value of vouchers. Not just difficult to present as an algorithm: this one is marked as “not going to be amenable to analysis at all”. It’s one of those odd limits that isn’t about CPU power or human intelligence.
So when I received an invite to the University of Eindhoven, to take a look at its hybrid project to evaluate the aerodynamics of a group of racing cyclists (or peloton), I was genuinely open-minded. This could be a cycle nerd thing, a wind tunnel thing, or a complete red herring thing. As it turned out, it was much more interesting than any of those, because it’s a supercomputer thing.
The question posed by Professor Bert Blocken was easy to understand. Cyclists are aerodynamically messy, with all those flapping limbs, whirling pedals and spoked wheels. Cyclists know this, and development of funnylooking carbon fibre egg-like fairings goes on apace to solve the physical problem. However, in pro cycling, the regulator is king, and the regulator
says leave the draggy naked-bike shape and the rider as they are. This has resulted in the creation of the peloton, because riders and team managers have been reasonably sure – on an empirically assessed basis – that in the middle you get much help from the slipstream of the riders around you. How much, exactly? The sport says 50%, maybe.
The approach taken at the University of Eindhoven was to treat this as a computing job: the relevant discipline is called CFD, or computational fluid dynamics. Much beloved by those with very large computers to sell, this is the field where the more power you can deploy, the more likely it is that your results are accurate. This makes the software you use curiously detached from the precise nature of the problem. You can model the flow of oil in a tube with the same product that Professor Blocken and his team used in this simulation. How did they do it?
Easily described, again. The lead PhD researcher sat on his racing bike, and the researchers 3D-scanned him and the bike. Then they laid out that model in the virtual wind tunnel of the simulation, making 121 copies and laying them out in a peloton-like formation. CFD’s consistency of approach across disciplines then governed their decisions over how many data points they wanted. That is, how many specifically designed volumes of air they wanted to track, to make up their model. You can see the 2D interpretation of these regions in the picture of the anonymous cyclist’s helmet and face on this page. As a general rule, the closer to the surface of the object, the smaller the cell used by the model.
So let’s see what this does to my point about people mis-estimating the scale of a computing project that models or receives data from the real world. 121 virtual cyclists, all identical. How many cells were in the resulting CFD model? No cheating. See what your intuition leads you to. A few thousand per model, times 121… can’t be that big, right? The answer is, three billion. This is currently the world’s largest computational fluid dynamics model, requiring 54 hours of runtime on a Cray supercomputer the size of a music festival toilet.
The assembled journalists were possibly somewhat heat-stroked at the start of the summer of 2018, because the questions floated to and fro, between the rules and habits of competitive cycling and the metaphysical limits to computation. The most relevant question was from a sports writer, who pointed out that cycle racing isn’t so regulated that all the riders are identical, and with the figures coming out of both the model and the matching wind-tunnel tests, this might be enough to neutralise the proposed gains. Whirling feet and legs, different physical statures, bike design: each factor is worth a few per cent. The riposte was that both the model and the real-world data point to expending only 10% of the effort required if you’re in the middle, at the back of the peloton, compared to riding on your own.
This might mean something for cycle race freaks. I confess that while this seemed like a revelation to the sports reporters in the room, I was still boggling over the IT project part of the story. Mostly my astonishment was in the horsepower-to-findings ratio. Two full days on a Cray with a decent amount of connected storage is more time than it takes to do a weather forecast for a large swathe of the Earth’s surface.
I’m sure that Cray would demur and point out that there’s a lot of customisation of the machines for each job. I don’t think that reduces my point at all, due to the opposing, simplifying force and universal nature of Ansys – the CFD modelling tool used in the project. You could approximate the model with much fewer points on much smaller hardware, but then you wouldn’t get the attention from the CFD community in academia and business. And no doubt there was a bit of grandstanding here – because CFD is a field where the same basic maths can be re-applied in different models of reality. If you get yourself a reputation in one field, you can easily apply what you’ve learned to others.
This is a pretty consistent trend in supercomputing. When you look around at the bigger players, you find they’re bursting with pet projects. Cray sent a couple of people to the press conference in Eindhoven, because while its normal stomping ground is weather forecasting and analytics out on the edge of what’s possible, it still makes sense for the company to show an interest in more esoteric fields of research and computation. A discussion about cycle racing spreads across businesses, which otherwise would be terrified of losing their competitive edge if they even said a word about their own internal modelling projects.
If you’re a guy in a business trying to make a product work better with materials or performance modelling, then I know this looks like an unattainable, extreme, academic exercise with no relevance to your business, problem or, indeed, budget.
For me, there are two takeaways here. The first is that initial estimates are remarkable mainly for their inaccuracy, and this project is a great thought experiment to put in front of those who might not stay awake through the complexities of a fully detailed, business-grade modelling presentation. The other is that the limits to maths are far closer than we think (so there’s proof of the initial estimate problem, too), and therefore the limits to computing are still going to be a problem – and a field rich with opportunities for the forseeable future.
More unsung heroes
Let’s hear it for the older Mac aficionados. It’s hard to recall how productive people could be,
“Network traffic analysis is like taking a cool drink from a blasting fire-hose”
especially in the design world, working on Macs whose CPUs would these days come bottom of a comparative review of smartphones. I still have a Mac PowerPC tower sitting in the basement, saved because it has installs of all the mainstay applications from back in the day: Quark, InDesign, Photoshop, Illustrator... The older versions aren’t licensed like the modern ones, which makes the temptation to keep those old fossils staggering along rather too strong to be denied.
There’s very little reliable information on how long such a machine should last, and what you can do to help it. On the basis of the response I’ve had to my recent “Unsung Heroes” roundup on Windows utilities, and the equally surprising longevity of a piece I wrote on alphr.com on how to revive your ancient iPod, here are a few top efforts for Mac users who want to extend the life of their hardware.
First off, disk duplicators. You can’t be in the life-extension business and be on your original hard disk, and the Mac is no exception. There are two contenders: SuperDuper! ( shirtpocket.com) and Carbon Copy Cloner ( bombich.com). Those who are students of the software business will have noticed that these two lifelong competitors share a somewhat cantankerous attitude to meaningful website domain names. Also, they share an odd blind spot, in that neither explains the way Macs format and partition their disks and how this affects the backup and restore process.
This isn’t necessarily about GUID partitions: it’s the general observation that you can’t fully clean up an ex-PC hard disk for use in a Mac, new or old. This applies across SSDs, laptop drives, even the hybrid SSD/spinning types that get called “Fusion drives” by Mac types. They all have to be cleaned before they can be used – by a PC. I believe this tiny gap in the toolbox on Macs is the genesis of the first wave of malware cleanup utilities, because messing about with a few bytes in the boot blocks of a hard disk ought to have been fixed 10 to 15 years ago.
But it hasn’t been. Neither backup utility lets you handle the problem inside its own menus, which drives people to look for any solution that isn’t typing “clean” into the command line interface to the Windows DiskPart utility on some mate’s PC.
Sadly, there are also departures from this sector in areas that remain useful and will be sorely missed. I was a fee-paying customer of Coriolis Systems, mostly for iDefrag and iRamDisk – two utilities that definitely rescue an Apple machine that’s used every day, but which plainly have suffered from lack of upkeep themselves. If you look for Coriolis now, you’ll see that the firm is concentrating on audio enhancers. iDefrag is at least still findable.
Simple downloads such as these set expectations among users, and by doing so opened the door to unprincipled developers, who majored on snooping around your machine, but didn’t do too much to actually help you. While the App Store in macOS was an almost-immediate solution to malware overnight, that doesn’t help those trying to keep ancient Macs staggering on. Many of the machines I see are restricted from running a late enough release of macOS to give access to the App Store. So you have to take a look at the traditional resources for finding helpful utilities. And this month at least, Google is showing up a new player on the block: macpaw.com.
Trying MacPaw’s cleaner app was a bit of a blast from the past. It folds together several older utilities actions. In particular, MacPaw gets rid of languages you aren’t using, which is a straight copy of Monolingual. PC users are a tad incredulous when shown how much space of a Mac boot disk is devoted to unused language files. Combining a run of Monolingual and a defragmentation utility could produce remarkable improvements on an ancient Mac. MacPaw’s features seem to be closely aligned to that whole ancient machine experience. It knows about the spread of junk files that arise from years of use, and makes reassuring statements about cleaning up after apps are removed rather better than the standard processes permit.
But I tried MacPaw on my old machine – a Mac Pro 2.1 with two four-core Xeons, courtesy of a decommissioned HP server and an Nvidia Quadro graphics card – and almost immediately took it off again. The reason is one of those problems that actually dogs everybody in the modern world: actually verifying the cause of a problem. In my case, after using MacPaw, my machine displayed a tendency to slowly but surely ramp up the graphics card fans as the day went on. I’d taken advantage of the fact that many appropriately aged Nvidia cards sold for PCs include Mac-capable firmware, too, and so my old machine is quite souped-up for graphics. That means double fans and a well-capable heatsink.
Post MacPaw, those fans have been putting in a lot of work. Slowly rising graphics card heat is a sign of the modern scourge of Bitcoin mining, embedded within some component you’ve downloaded. If I’d tried the application on the “Upstairs mac”, which is some three generations later, the greater efficiency of the graphics card would very likely have masked the effect entirely.
My problem is, old or new, unsung or otherwise, it’s remarkably difficult to get good traceable trails of infection out of any machine these days. I don’t want to have to start monitoring the network traffic of the machine, despite having a hardware LAN tap and the relevant install of Wireshark to hand, because emulating how a regular single-machine user could resolve this dilemma is part of my brief. I’d like to say I’ve found a new breed of unsung hero but, as we go to press, I can’t say for sure that MacPaw is the source of my problem or the resolution to it, after a few more rounds of cleaning and de-junking.
Despite the moves made by Apple to resolve this kind of diagnosis gap, with an App Store and digital signatures, it seems to me that life for the older Mac aficionado hasn’t been getting any easier. cassidy@well.com
“You can’t be in the life-extension business and be on your original hard disk – and the Mac is no exception”