Interesting Links for 05-09-2018

[andrewducker]

Reboot Your Aeroplane Every 248 Days To Avoid Crashes

(tags: airplanes software OhForFucksSake )

Probably the finest headline of all time

(tags: headline journalism viaPatrickHadfield )

Why computer systems are so complex and expensive

(tags: computers design )

Radical open-access plan could make research instantly available to all

(tags: research OpenAccess science GoodNews )

Reoffending rates fall in Scotland

(tags: crime scotland GoodNews )

Massacre in Myanmar: One grave for 10 Rohingya men

(tags: genocide Myanmar )

In A New Email, Elon Musk Accused A Cave Rescuer Of Being A “Child Rapist” And Said He “Hopes” There's A Lawsuit

(tags: ElonMusk wtf )

Four million UK children too poor to have a healthy diet

(tags: food UK poverty OhForFucksSake )

Michel Barnier proposes Canada-style Brexit deal over May's Chequers plan in meeting with MPs

(tags: uk Europe doom )

Super altruists (who’ve donated a kidney to a stranger) show heightened empathic brain activity when witnessing strangers in pain

(tags: empathy psychology )

Comments

[drplokta]

Tha aeroplane article is rather disparaging of increasing the number of bits in the counter as a solution, just saying "it puts off the overflow". Yes, using a 64-bit integer rather than 32-bit would "put off the overflow" from 248 days to 30 billion years, which seems like a real solution to me, not just a quick and dirty fix.

[andrewducker]

It does rather. 30,000,000,000 years ought to be enough for everybody!

[drplokta]

Ah no, I've misplaced a decimal point. 3 billion years, which should still be more than ample.

[doug]

Yeah, I'm convinced that would address the immediate issue, but I would be astonished if switching to 64-bit ints was a simple fix. Even in ordinary, simple corporate stuff you can get bitten by weird problems when you change types like that - implicit conversions go a different way, library functions misbehave, APIs and other software interfaces don't work (or, worse, silently discard excess bits), and so on. Much of that you can deal with at compile time, of course, especially if you have effective tests. (Although you may well also need to update your tests.)

But this is in a high-reliability embedded system, which raises the stakes and the constraints, and it's almost certainly in a real-time part of it (why else would you need to keep track of time with 10 ms resolution?), which does the same squared. Tight space and time constraints are not a fun mix with doubling storage requirements.

Without knowing the implementation details, one imagines that a fix is almost certainly do-able, of course - but one can also easily imagine that the effort to do so is hopelessly uneconomic compared to the cost of simply rebooting every plane every 120 days after it was last certified to be powered-down. (A smart choice of interval - gives you a bit of margin for deferred maintenance even if they missed it off the list last time.) The cost of compliance is estimated at $85 per 'deactivation cycle', which as airline maintenance things go is not a big ticket item. And it looks a lot like this particular maintenance action is vanishingly unlikely to ever be triggered anyway, since planes are required to be powered down for other scheduled maintenance more frequently than every 120 days.

I love aviation safety culture.