Hacker News Reader

Comment by megamindbrian2 | Hacker News Reader

I see these issues everywhere also, it is really worrying. Especially since our entire food and supply chain is starting to depend on it.

View on HN

There is no money to be made in making systems more robust, not until problem really arises. This worries me great deal.

Is there a reason they have to be "more robust"?

Lest, e.g. some Amazon deliveries don't come in 1 day but take a week?

Or some employee doing some BS job can't use Gmail for a couple of days?

Except the failure mode for Amazon shipping software (in this example) is equally or more likely to be the item is never shipped or a completely different item is shipped, rather than the correct item being delayed. The latter two are more difficult to deal with. Also, even in the delay case, more people are depending on Amazon for items they truly need, like groceries. Delays can matter.

Supermarket perishables are the classic example of things that have to be delivered within a very short time. Earlier this year there was no fresh milk, bread, veg or meat in my local (urban!) supermarket for a few days after some snow.

>Supermarket perishables are the classic example of things that have to be delivered within a very short time.

Yeah, but we've managed that in the 60s and 80s without so much as the web. Trucks are still needed of course.

But that doesn't imply that the computer system can fail over to paper overnight, or that it's cheap to do so.

Yes, in some cases.

Like thermostats. I've read about Nest thermostat crash on loss of Internet connection which result in dangerous temperature loss in infants room. We can make those thinks more robust for years, but it does not have an app and does not make so much money.

Food supply was original comment. Most people do not have food for week. (I compare it to always stocked pantry of my grandparents). Any small - one week - disruption can result in big problems for some.

But we are moving to hi-tech just in time system everywhere, which are dependent on Internet, GPS, advanced forecasting etc. Those systems can be for example hacked and not functional for days.

I don't care about Amazon delivery being late.

The dangerous temperature loss from a Nest thermostat crash sounds like hyperbole.

https://www.nytimes.com/2016/01/14/fashion/nest-thermostat-g...

That's what I mean, 64 degrees isn't the end of the world. And for the few hours of the night an infant sleeps, it's the same issue as if the power/gas were to go out.

How about in Arizona where it is 120. Too hot could kill someone easily.

A power outage results in the same thing. Someone so incapacitated that they can't move from a place that's getting too hot (such as a baby or the infirm) aren't probably left alone for long enough for the temp to reach extremes in the first place, and if they are, then that's the actual "dangerous" problem, not the thermostat.

I'm disagreeing with, or pointing out, the incorrect use of the word "dangerous" in order to get clicks for an article.

It may sound like hyperbole, but there are contemporaneous reports of real people noticing their rooms were rather cold despite their settings.

That might be, but I wouldn't use the word "dangerous" to describe the effects of a thermostat not working. It's the same as when the power goes out. Obviously, no one is leaving their baby alone for hours and hours and hours while the temp actually gets to freezing level such that it would harm the baby, so what is so dangerous about it?

I can see the harm in frozen pipes and whatnot, but you take the same risk when dealing with other systems, such as other thermostats or power or the furnace itself. However, if it is known that Nest devices fail when the internet fails, then that is Nest's responsibility to advertise that to users.

I've never seen a Nest that worked the way it was supposed to. The last time I took a look at the one in the office, it had a bizarre set of temperature set-points bouncing around up and down 20-30 degrees F every 5-7 minutes. I have no idea how it got into that state, but it's no wonder that it doesn't work worth a shit. We've also had issues where it was linked to a cellphone app, and got incredibly confused when that cellphone went on vacation to a different timezone...

The best thing to do with them, and most of this other godawful "smart" home tech is to rip it out, smash it with a hammer, and put back the cheap analog versions you had in originally.

It tries to detect when you are home/out and adapt the temperature settings. Only when it's in ECO mode will it not change the settings.

Looking for a Wifi connection will drain the Nest's battery faster than it can recharge. It trickle charges from the "power on" thermostat lines.

But mine has not outright crashed when its Internet connection was interrupted.

> Or some employee doing some BS job can't use Gmail for a couple of days?

I'll grant you Amazon, but turn off GMail for a couple of days, and you'd dip the economy. A great many of small but essential companies - from car workshops to paint shops to furniture makers to retail and wholesale shops - run their business on personal GMail accounts. Losing access to primary e-mail for a week may very well be an existential threat for them. One which they don't realize, because GMail is generally reliable.

Just anecdata, but 2 years ago Amazon incorrectly shipped me a 90lb granite surface plate instead of a 1lb set of calipers. Return shipping was ludicrous enough that customer service offered a ~90% discount if I elected to purchase it. So a software snafu (because I doubt a warehouse worker would ship an item that heavy out of laziness) cost them significant money, and extrapolate that cost by the number of orders daily: a serious glitch could run to millions of dollars in short order, and have long-term loss of trust issues for the business.

> There is no money to be made in making systems more robust, not until problem really arises. This worries me great deal.

You say that but big companies like Amazon can calculate how much money they're missing per second of their services not being available, or how shaving off x milliseconds off of each request reduced their cost by millions per day. They also advertise with their uptime guarantees to haul in big contracts.

Amazon won't commit to uptime beyond refunding you any charges when they're down. This is insufficient for critical infrastructure.

Disclaimer: I am in risk management for financial infrastructure.

Luckily, there's no giant central food-supply IT system, which makes the food supply as a whole fairly fault-tolerant.

If there's anything that could make the software better, it has to be an economical incentive. Learning to build reliable real-world systems that rely on faulty software (you might call it IRL erlang) would be great, too.

I see these issues everywhere also

I know you probably mean this for software, but it's not limited to just that, and it's as ancient as humans are (or animals, to take it further, they make mistakes as well from time to time). I mean, instead of booking.com messing up (or whatever it was) I've already arrived at a hotel once just to find out there was no trace of my booking. Someone lost the paper or forgot to write it down or so. Turning the human off an on again, a.k.a sleep, also helps with this from time to time and makes it a nasty hard to reproduce bug. And hard to fix as well. I get the analogy only holds - usually - when it's not about bugs which are reproducible time and time again. But on the other hand those are usually easier to discover and fix so maybe less of a problem. tldr; not sure how worrying it is.

Some designer wrote (Russian, of course, we're all into this stuff) something along the lines of "If you feel an area is great — you're just not familiar with it". Tried not to get carried out there, I'm already sad enough)

I guess (hope) avionic / nuclear software is generally of higher quality, but must it really be so hard and expensive to develop well?

Yes! That's the issue. Good, robust development takes time. You can estimate the number of flaws in a product by the amount of time it takes to find one - if this is measured in minutes, your product likely has thousands of errors.

Actual, robust software takes thousands of hours of TESTING, nevermind development. Multiple vendor cycles. Field studies. User observation to limit system error.

And unfortunately (if you're not a programmer) programmers are very, very expensive, but most software must be very cheap. So unless you're at the largest scale, your software is likely crap or very expensive, because the market won't support anything else.

Reply on news.ycombinator.com