Posts by David Gillies

Doesn't add up

Petrol has an energy density of 34.2 MJ/l. Flow rate on a US petrol pump is limited to 38 litres a minute, but even so that's delivering 1.3 GJ a minute or 21 MW.

Don't go exploring in academic codebases without a pump-action shotgun and a torch

The overall quality of code in academic STEM disciplines is hair-raisingly awful. I used to get called in on occasion to help out a colleague with a particularly refractory bit of code. I think my greatest win was applying a bit of algorithmics and dropping the complexity of a routine from quintic to linearithmic which in the problem space where it was running represented five orders of magnitude speed-up (which is a month to 25 seconds).

Tarquins and Jemimas, the lot of 'em

Sod off, Swampy.

There are well-known threshold schemes for secret sharing such that a key is split into, say, eight pieces, any five of which can be used to reconstitute the key. Fewer than the threshold and the secret is irrecoverable. I've even got some C++ code knocking about that implements this with nim multiplication over GF(256). I dunno what sort of shonky operation has a single point of failure like this.

And seriously, if you've got inflammatory bowel disease, is a holiday in India good idea?

Some things are forgivable

But bringing down a mission-critical resource because you've forgotten to renew the license is not one of them (q.v. the recent O2/Ericsson outage; I would hope multiple P45s ensued).

Magical thinking

It's legal to make guns in the US. Always has been. You can build your own 3-axis CNC machine for $1000 or so, and mill a lower receiver or a trigger group to a few tens of microns tolerance. It's well within the capabilities of a hobby machinist to rifle his own barrels. Steel of a suitable grade for a gun barrel is available mail order. There are hundreds of people with maker channels on YouTube who have the requisite skillset to manufacture a firearm, which means thousands or tens of thousands total. How many guns are made by hobbyists a year? Unknown, but certainly a lot. How many have been used in the commission of a crime? If the number isn't zero, it's as close to zero as to make no difference. The intersection of the set of people who make guns and the set of people who use them to commit crimes is the empty set, as near as dammit. The people who make Viking battleaxes and two foot-long Bowie knives in their home forges aren't driving round on mopeds stabbing people.

Lawyers and politicians trying to stop hobby firearms manufacture are either ignorant or disingenuous. Claiming that gun crime will be lowered if gun blueprints are restricted is magical thinking.

When Flash is finally retired they need to seal the source code in an Inconel block, with suitable biohazard stickers, and get one of Elon Musk's BFRs to fire it into the Sun.

"Mansplaining"

Complaining about men explaining things on a site devoted to explaining things to people working in a male-heavy field seems...curious. Actually, "wilfully perverse" is nearer the mark.

Standards are great; that's why there are so many of them

It was a retrograde step allowing Unicode in variable names, never mind this.

Reality is that which, when you stop believing in it, doesn't go away

Huzzah. Another instance of politicians behaving like a cross between King Canute and Humpty Dumpty. I suppose to a politician all things must have a political/legislative solution, otherwise what is the point of their existence? But really this is tantamount to the attempt in Indiana to legislate a method for squaring the circle. Given that the set (World Population) \ (UK Population) is approximately equal to the set (Everyone), what exactly do Theresa May and her cyber-myrmidons hope to achieve? Clearly the denizens of Italy or New Zealand or Chile are not going to be thwarted by UK age verification in their quest for a crafty J. Arthur, so how, in the name of all that is holy, is this supposed to work? Have any implementation details been made public beyond a vague statement of intent?

Linux version has been crap for weeks

The latest 8.14 release of skypeforlinux is completely non-functional, and it appears 8.12 and 8.13 don't work either. It shows nothing but a blank screen. This is on a generic Ubuntu 16.04 installation. I had to downgrade to 8.11 to get it working. Microsoft's QA is generally pretty abysmal but this takes the cake. People have been shrieking on the laughingly-titled 'support' forum for ages and the response from MS has been the square root of bugger all.

If you eat foodstuff A in lieu of foodstuff B then foodstuff A _is_ a meal replacement, just as B would be were it eaten rather than A. If people want to consume some vile tofu concoction rather than something wholesome then that's their lookout. Unless it's actually toxic, governments have no remit to poke their noses in.

How does "the inside of your car" equate to "a public space"? And event if it is, why is screaming in public spaces not allowed? And haven't the Dudley Do-Rights got something a teensy bit more urgent to be doing?

I'm glad we're telling them about the flaws in their cryptography. It reminds me of that press conference we had in 1943 when we told the Germans we'd broken Enigma.

If you're writing for a living and don't know what 'intransitive' or 'gerund' mean then you should probably have your thumbs broken, to stop you writing sub-literate gibberish.

It doesn't work on Linux without a lot of grunting and squealing so it's de facto broken. Dropbox works like a charm.

That pseudocode's got a buffer overflow error in it.

FOBS

We can rule out a terrorist organisation doing this, but the threat of a State actor deploying an EMP weapon is not an empty one. It might not be top of the list of threats, but Fractional Orbital Bombardment Systems (FOBS) have been known about for a long time. These have a strong deniability component and are also very difficult to counter. Civilian infrastructure would be very badly impacted. For example, food distribution is heavily reliant on just-in-time delivery and refrigeration, both of which would be badly hit by a significant EMP event. You don't need a multi-megaton weapon to cause a non-negligible EMP. In fact fission weapons give you more bang for your buck if you want to take out electronic infrastructure like telecomms and computer equipment. A big thermonuke will be better for knocking out power distribution systems. Both are hard to safeguard against.

It's easy just to dismiss this as lobbyists angling for cash, but it's a real problem.

@Dave

Rorke's Drift a Pyrrhic (note spelling) victory? Not hardly. 17 killed out of 139. The idea that the braver side always wins is rubbish. Who were braver at Omdurman, Kitchener's army or the Dervishes? Who were braver on Iwo Jima, the Marines or the Japanese defenders? Courage is a very important factor, but technology can produce lopsided results irrespective of the relative bravery of the combatants. It took just as much courage to fly a bomber over Germany in 1941 as in 1945, but in 1941 crews were lucky to deliver their bombload within five miles of the target (frequently they bombed the wrong country, never mind the wrong city). By war's end average targeting error was down to 300 yards, with some specialist units achieving as little as 80 yards. The difference in combat effectiveness was profound, and it was almost wholly due to technology.

I'm also a bit fed up with the assertion that US troops are mindless undeducated automatons given to panicking and shooting indiscriminately. Nothing could be further from the truth. Their NCOs, in particular, are absolutely excellent, and it is not uncommon to find, say, a SSG with a Bachelor's or even Master's degree.

I'd be tempted to call bollocks but...

...one of our guys has a six month old 64-bit AMD machine with 2G of RAM and simply task switching from, say, Photoshop to Zend Studio can take up to 30 seconds a time. I'd be a gibbering homicidal maniac after an hour or so of that nonsense. His machine is, no kidding, about a tenth the speed of the box I'm writing this on, which is a five year old 1500 MHz PIV with a gig of RAM. But then I'm running Ubuntu Dapper. I still use Mandrake 9.2 at work on an identical box and it's a sight faster. At least 25 seconds of the boot cycle is firing up the Adaptec Ultra160 SCSI card. A pure ATA configuration is POST to XDM login in 45 seconds.

@Chris C

Yeah, it's a bad idea to give external devices truly direct access to memory. So map it through virtual memory. Or you can disable the OHCI hardware mapping, albeit with an efficiency penalty. Of course if you're close enough to a machine to be plugging things into it, security is somewhat moot (unless someone cons you into hooking up a trojan device, but that's a bit too Mission Impossible to be plausible).

The real problem with USB vs 1394 is that USB is master/slave and 1394 is P2P. Unless that's going away, and I can't see how it would, real world 1394 performance is always going to exceed USB at a given nominal data rate. What does USB 2.0 really give you? Around 40-45% of nominal bandwidth in my experience. Firewire pledges to give you 97%. 45% of 4800 Mbits/s is less than 97% of 3200 Mbits/s.

Body bags?

You didn't get 'sent home' in a body bag or a coffin or a bit of blotting paper in an envelope. You were buried in theatre. The second largest US military cemetery outside Arlington, VA, is in the Philippines.

If you try to buck the market...

...the market will buck you.

Saturating my relatively thin cable connection works out to about 2/3 TB per month. That's unrealistic, but I usually run through a tenth of that at least, or 60Gb/month. Someone has to pay for that infrastructure, and that investment has to return at a rate that makes it viable. This is really quite simple: no-one in their right mind is going to invest umpty-squillion simoleons in something if it exceeds the present value of the revenue stream they rationally expect to derive from it. It's all very well crying about the fat cats capping your bandwidth, but the money has to come from somewhere. Trunk bandwidth is relatively easy to scale, but deploying fibre-to-the-kerb is very, very expensive (i.e. on the order of $100 per metre to dig the road up, lay cable, and reconstitute the surface). Switching hardware is not cheap either, and it's a constant battle of Moore's Law lowering of hardware costs vs. Parkinson's Law of ever-increasing bandwidth demand. There isn't a magic bandwidth fairy that pays for all this. Either it's paid for through government subsidy (taxes) or directly via the consumer (higher bandwidth costs) or the bandwidth you're quoted is not available (caps/contention). I fully agree that the terms of service should be as explicit as possible in detailing what the customer will receive. Having said that, the dominant driver in lower price-per-gigabyte is a strongly (and truly) competitive telecoms market, and that is one area where the US is well and truly shafted. You really want cheap broadband? Abolish the FCC.