Super long-term kernel support

Posted Mar 19, 2018 20:04 UTC (Mon) by Cyberax (✭ supporter ✭, #52523)
In reply to: Super long-term kernel support by arjan
Parent article: Super long-term kernel support

A lot of civil infrastructure quite explicitly avoids upgrades. A lot of utilities, for example, don't do incremental upgrades at all and instead do a forklift migration every 20 years or so.

It was not uncommon for utilities to switch from green-screen VT100 terminals running over X.25 straight to LCD displays and fiber-optic TCP connectivity in their control centers.

This is reasonable, considering that risks of downtime are extremely high. You can orchestrate one migration every 20 years, which usually includes two sets of control centers running at the same time and doing the handover. But doing this on a constant basis is not sustainable.

Super long-term kernel support

Posted Mar 19, 2018 21:17 UTC (Mon) by atelszewski (guest, #111673) [Link] (17 responses)

Hi,

There goes this saying: "Never touch a running system".

--
Best regards,
Andrzej Telszewski

Super long-term kernel support

Posted Mar 19, 2018 22:04 UTC (Mon) by arjan (subscriber, #36785) [Link] (8 responses)

.... except for security updates

and maybe a feature or two

and .. and ..

and then your backports are high risk since the code now runs in a context it was never tested before

Super long-term kernel support

Posted Mar 19, 2018 22:48 UTC (Mon) by tlamp (subscriber, #108540) [Link] (5 responses)

> and then your backports are high risk since the code now runs in a context it was never tested before

the same could be said about a new kernel, it contains thousands of lines not tested in the environment needed...

Easier to ensure one really needed security feature gets back ported right, when the need arises maybe once in 10 years, then a whole control system underlying kernel gets just swapped out every few weeks....

And no, your ordinary civil infrastructure project doesn't needs the newest fancy syscall, IO scheduler, whatever feature, at the moment it gets released.

As other said *never* touch a running system, this is not about a a small daemon or web app of yours, this can affect millions of people and whole economies in a meaningful way!

Super long-term kernel support

Posted Mar 20, 2018 16:32 UTC (Tue) by mjthayer (guest, #39183) [Link] (4 responses)

Just on a devil's advocate sort of line of thought, what would have to happen to make the mainline kernel usable for this sort of purpose? Not that I can imagine anything that would make either side (the kernel developers and maintainers or the consumers) remotely happy.

Super long-term kernel support

Posted Mar 20, 2018 19:48 UTC (Tue) by smoogen (subscriber, #97) [Link] (3 responses)

For the mainline kernel, it would require that a kernel was not released until it ran through every different industries test suites which would have to be run on a wildly different hardware and environments (the same system may need to be tested in -40F, 0F, 32F, 120F, 180F over N days depending on what environment the system is supposed to go into.) It might need to be run in a turbine at different speeds while other environment factors are going. Etc etc. It would also require that all those industries publish their testing systems and make them available to developers. [Airplane radars, satellite communication, medical equipment, manufacturing devices, ATMs, sewage control, water pumps, oil pumps, fire control devices, traffic signal devices, etc all have different testing requirements that depend on the state/province, country, and treaty area. Some of them may be short done testing and others require the entire environment be replicated and run from the ground up until the device has failed.]

For a long term security kernel, it would take them to wait the 6-9 months for feedback from one set of changes to be run through.

These devices are going to sit on a shelf for years at a time until put into replacement due to some forklift upgrade. They will then get looked at years later. Most of the devices may be only hooked up to some sort of serial network so updates are done by hand as the bandwidth for updating is faster that way.

Super long-term kernel support

Posted Mar 21, 2018 12:01 UTC (Wed) by mjthayer (guest, #39183) [Link] (2 responses)

And presumably then the sort of updates such a SLTS kernel will receive are not going to include fixes for any security problems which can be solved by better isolating the device running it (which will also presumably not be in any way network-facing)? So no Spectre and Meltdown for example. Only conservative bug fixes to prevent crashes which might occur in the normal course of operation.

Super long-term kernel support

Posted Mar 21, 2018 14:10 UTC (Wed) by mjthayer (guest, #39183) [Link] (1 responses)

Ahem, I could have read Cyberax's comment slightly further down.

Super long-term kernel support

Posted Mar 21, 2018 14:19 UTC (Wed) by mjthayer (guest, #39183) [Link]

That said, they say that "CIP works towards reducing the window of vulnerability to zero".

https://wiki.linuxfoundation.org/civilinfrastructureplatf...

Super long-term kernel support

Posted Mar 20, 2018 7:56 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link] (1 responses)

No. They don't do security updates or new features, unless they REALLY have to.

Security provided by network separation and features are usually provided by new systems working in parallel with old systems.

Some utilities even stockpile hardware, so that they can replace failing components with hardware from the same batch.

Super long-term kernel support

Posted Mar 20, 2018 9:13 UTC (Tue) by Mog (subscriber, #29529) [Link]

Not just utilities. A few years before Y2K I visited a client, a printer, and they showed me a dozen Sun3 (yes 3) stored in a room. They had some software running on SunOS 3.5 that they planned to replace in the next decade or two but for the time being, they needed it running. Their Y2K fix : switch the system clock 28 years back and a bit of scripting.

Super long-term kernel support

Posted Mar 20, 2018 13:02 UTC (Tue) by arjan (subscriber, #36785) [Link]

also obviously.. if you're never going to update at all you don't need an LTS kernel.
An LTS kernel, with the S for support which implies maintenance (such as security) you will get more changes than your specific use of it would strictly need.

Super long-term kernel support

Posted Mar 20, 2018 21:26 UTC (Tue) by JFlorian (guest, #49650) [Link] (6 responses)

I much prefer a "touch all the time policy". Sure you'll get fallout. So get better a dealing with it. It's going to happen eventually anyway.

Super long-term kernel support

Posted Mar 21, 2018 12:04 UTC (Wed) by mjthayer (guest, #39183) [Link]

You might still prefer to have as tight control as possible over how and when the fallout will occur, and to minimise it as much as you can. If the fallout can include loss of life, just "touching" and watching to see what happens might not be the best way to handle it.

Super long-term kernel support

Posted Mar 21, 2018 17:44 UTC (Wed) by Cyberax (✭ supporter ✭, #52523) [Link] (2 responses)

Congratulations! Your neighborhood has been randomly chosen to participate in a staging area for Linux 5.11.3 deployment. In case of electricity blackout or water mains rupture, please relocate to another area. Also remember to stay away from all wires.

We are very sorry for the last time we connected a 50kV line instead of a 5kV, we're pretty sure it doesn't happen with this patch.

Thanks!

Super long-term kernel support

Posted Mar 21, 2018 17:53 UTC (Wed) by JFlorian (guest, #49650) [Link] (1 responses)

There's a very large difference between intelligently managed continuous updates and the reckless haphazard scenario you seem to think I'm prescribing.

Super long-term kernel support

Posted Mar 21, 2018 18:04 UTC (Wed) by Cyberax (✭ supporter ✭, #52523) [Link]

Well, most of industrial hardware designers have processes for responsible updates.

They usually involve months (if not years) of testing both software and hardware under varying conditions. You wouldn't want your thermal management system to freeze because an update introduced a subtle memory leak or a race condition that become apparent only after a couple of months of runtime.

CIP will provide a better foundation for it, but it most definitely won't solve the issue of long deployment cycles.

Super long-term kernel support

Posted Mar 29, 2018 12:07 UTC (Thu) by federico3 (guest, #101963) [Link]

This is what actually happens in industrial automation, banking, airplanes and phone carriers.
Everything gets upgraded all the time but, obviously, there are layers of safeguards.

Source: I work on CI/CD/orchestration systems that are used in those fields.

Super long-term kernel support

Posted Mar 29, 2018 19:21 UTC (Thu) by Wol (subscriber, #4433) [Link]

> I much prefer a "touch all the time policy". Sure you'll get fallout. So get better a dealing with it. It's going to happen eventually anyway.

"If it ain't broke, fix it till it is" :-)

If you're talking hardware that's expensive, you're not going to replace it. Why UPGRADE the kernel, when the majority of changes is adding new hardware drivers, if your system has no new hardware that needs it? And if it's not new drivers, the rest of the new code is pretty much equally useless FOR YOU.

Think about the story of when they dropped serial ports from hardware. Apparently Bill Gates' response to one customer's complaints was "well, buy new peripherals, then". 20 industrial machines, at $250K apiece??? All for the sake of $10 board in a computer?

Cheers,
Wol

Super long-term kernel support

Posted Mar 19, 2018 22:28 UTC (Mon) by rahvin (guest, #16953) [Link] (31 responses)

Almost all the new advanced car systems run on Linux and it's not uncommon to have cars 20+ years. I still see pickups from the 70's on the road. Industrial controls and appliance computing typically has a life as long as the factory, up to 20 years in most cases. There are endless examples where you have specific computing instances running on Linux that's going to be expected to run for decades. I think the idea of a 20 year supported kernel are good ideas.

As time passes the number of exploits found and patched in those kernels should go down dramatically but fixing critical stuff like meltdown or heartbleed (not kernel based I know) should be ported back and these systems updated. Keep in mind that within a decade you may be driving around in car with Linux controlling all your safety systems if you aren't already, that alone should scare you if security exploits aren't being patched.

Super long-term kernel support

Posted Mar 20, 2018 10:52 UTC (Tue) by eru (subscriber, #2753) [Link] (30 responses)

I still see pickups from the 70's on the road.

Old Volvo sedans also seem immortal. But those did not contain any CPU:s and software, probably even the ignition control is electromechanical. They also were more repairable than modern cars. I seriously doubt that any "digitalized" automobile manufactured today can be seen on the road 30 years from now.

Super long-term kernel support

Posted Mar 20, 2018 12:39 UTC (Tue) by musicinmybrain (subscriber, #42780) [Link] (29 responses)

I have no doubt whatsoever that adequately maintained Honda Civics and Accords and Toyota Camrys from the mid to late 2000’s can still go 30 years or more. I am sure there are other such models, and even newer ones; I speak only of those I know well.

Super long-term kernel support

Posted Mar 20, 2018 13:00 UTC (Tue) by arjan (subscriber, #36785) [Link] (27 responses)

You might have a hard time getting fuel or parts in 20 to 30 years....

Super long-term kernel support

Posted Mar 20, 2018 13:19 UTC (Tue) by felixfix (subscriber, #242) [Link] (26 responses)

Peak oil as described ad nauseum in the hysterical media is a myth; there will still be plenty of fuel in such a short time frame.

3D printers will make parts will be easier to get than ever.

Super long-term kernel support

Posted Mar 20, 2018 13:36 UTC (Tue) by tao (subscriber, #17563) [Link]

I think few if any believe in peak oil, but I still hope that it'll be hard to get petroleum products 30 years from now, considering the bad impact fossil fuels have on the environment.

Super long-term kernel support

Posted Mar 20, 2018 14:59 UTC (Tue) by arjan (subscriber, #36785) [Link] (10 responses)

I was more assuming that as more and more cars go electric, gas stations will get more and more sparse.... the business model of a gas station won't last forever

Super long-term kernel support

Posted Mar 21, 2018 11:28 UTC (Wed) by jezuch (subscriber, #52988) [Link] (9 responses)

Some of them are already switching partly or fully into charging stations. This is huge infrastructure already in place, it would be foolish not to reuse it.

Super long-term kernel support - fuel availability for 20-year-old cars

Posted Mar 24, 2018 21:29 UTC (Sat) by giraffedata (guest, #1954) [Link] (8 responses)

Some of them are already switching partly or fully into charging stations. This is huge infrastructure already in place, it would be foolish not to reuse it.

How does a corner charging station work? Doesn't it take hours to charge a car?

I have a colleague who worked on battery technology and told me that a standard gas station nozzle delivers 30 megawatts, and that there was nothing on the horizon that could match that with electric battery storage.

Except that I read once about an idea for swapping out the entire battery.

Super long-term kernel support - fuel availability for 20-year-old cars

Posted Mar 25, 2018 4:55 UTC (Sun) by songmaster (subscriber, #1748) [Link]

Tesla built and demonstrated a machine that could replace the battery pack in one of their vehicles faster than you could refuel some high-end car. I’m sure there are YouTube videos of the press demonstration available. Elon Musk was talking about installing such machines around the country alongside their super chargers, but I think they must have dropped the idea since I haven’t heard anything about it recently.

I guess the energy capacity of a gas pump nozzle is somewhat analogous to the bandwidth of a truck full of hard drives driving down a highway — wires aren’t always the fastest way to transport energy/data.

Super long-term kernel support - fuel availability for 20-year-old cars

Posted Mar 25, 2018 15:52 UTC (Sun) by excors (subscriber, #95769) [Link] (3 responses)

I think a significant difference over petrol/gas is that you can (once the infrastructure is installed) charge your car when it's parked at home, at work, at the shops, etc. If it takes 8 hours to restore the energy you used on your commute, that's good enough. It's 2-3 orders of magnitude less than 30MW if you look at instantaneous power; but if you consider the amount of time you actively spend on it (driving to the gas station every week, vs plugging in a charger at home every day), electric probably has better joules-per-second-of-human-effort.

The high-power charging stations are for rare long trips. Apparently the Tesla Model S can get 170 miles of charge in 30 minutes. (Full charge takes disproportionately longer, so it's quicker to do multiple partial charges). So it's not hours, but long enough that I guess you'd typically want facilities (shops, food, etc) for people to use (and spend money in) while waiting. I guess with that, plus the reduced demand if most charging is done at home, it's not going to be able to support anywhere near as many charging stations as there are gas stations today.

There's also the Formula E approach where the drivers get a fully-charged battery in about ten seconds, by simply swapping their entire car. Not sure how well that would work with consumer vehicles though.

Super long-term kernel support - fuel availability for 20-year-old cars

Posted Mar 29, 2018 12:09 UTC (Thu) by NAR (subscriber, #1313) [Link] (1 responses)

"vs plugging in a charger at home every day"

I'm afraid many (maybe most) people don't have a home where they can plugin a cable from the car. Think about a place like this: https://goo.gl/maps/bgH6CUD7BBS2.

"The high-power charging stations are for rare long trips."

I guess cars currently spend about 3-5 minutes at the fuel pump: fill the car, go to the shop, pay, leave (maybe just to a parking slot). If cars need to spend about 10 times as much time at the plug, the motorway rest station will need 10 times more space - instead of 16 pumps, 160 parking places with plugs. Not sure all of them would have the place.

Super long-term kernel support - fuel availability for 20-year-old cars

Posted Mar 29, 2018 12:43 UTC (Thu) by jem (subscriber, #24231) [Link]

> I'm afraid many (maybe most) people don't have a home where they can plugin a cable from the car. Think about a place like this: https://goo.gl/maps/bgH6CUD7BBS2.

What's the problem? As demand grows, chargers will pop up in the car parks. I don't see why you can't eventually have enough chargers at the sides of the parking spaces to serve a 100 % electrified car fleet.

Super long-term kernel support - fuel availability for 20-year-old cars

Posted Mar 31, 2018 8:23 UTC (Sat) by daenzer (subscriber, #7050) [Link]

FWIW, the second generation Formula E car (used as of season 5, starting towards the end of this year) is said to have enough battery capacity that the car swap will no longer be necessary, despite delivering higher speed.

Super long-term kernel support - fuel availability for 20-year-old cars

Posted Mar 25, 2018 17:38 UTC (Sun) by Cyberax (✭ supporter ✭, #52523) [Link] (1 responses)

A regular (Level 2) charging station typically delivers around 2-6kW of power, so it takes several hours to charge a car completely. It's perfect for daily commute where you charge your car overnight at home or during daytime at work.

For fast charging, Tesla superchargers can deliver 150kW of power right now and 400kW chargers have been demonstrated. And you don't really need full 30MW for charging if you can drive 500 miles on a single charge.

Super long-term kernel support - fuel availability for 20-year-old cars

Posted Mar 28, 2018 3:08 UTC (Wed) by giraffedata (guest, #1954) [Link]

So, summing up: there is no practical way to convert a corner gas station to a charging station, so it probably is not happening today, and such conversions are not a reason to doubt that gas will be easy to find 20 years from now.

Super long-term kernel support - fuel availability for 20-year-old cars

Posted Mar 29, 2018 18:30 UTC (Thu) by Wol (subscriber, #4433) [Link]

> How does a corner charging station work? Doesn't it take hours to charge a car?

Full charge, or useful charge?

These figures are from when I was looking at possibly getting a Nissan Leaf about 3 years ago ...

Time to full charge - several hours.
Time to 80% charge - 30 minutes? A decent motorway services rest break.

So, bearing in mind our daughter lives over 200 miles away, with a range of about 200 miles we could get there *easily* with one short stop at a service station to recharge. Both the car, and ourselves :-)

And while the UK plans to ban the sale of non-electric cars by 2040, I suspect we will still have small liquid-fuel engines that can provide top-up power. I'm planning for my next car to be a mixed-mode car - primarily electric with backup petrol engine for when the range is insufficient.

Cheers,
Wol

Super long-term kernel support

Posted Mar 20, 2018 16:12 UTC (Tue) by nix (subscriber, #2304) [Link] (4 responses)

Peak oil doesn't mean we've *run out* of fuel: it just means the extraction rate has gone down and/or the amount considered extractable is declining.

I find it hard to believe that anyone would be so foolish as to consider that this would never happen. The Earth does not have an infinite volume, after all, and nearly all of it is nickel-iron and/or peridotite, not oil. Trivial back-of-the-envelope calculations show that we are using oil at roughly a million times its production rate, so eventually, if we keep extracting it, we will run out, unless we reduce the extraction rate more than a millionfold (which is obviously impossible without finding complete replacements for absolutely all oil's industrial uses, even the minor ones).

Super long-term kernel support

Posted Mar 20, 2018 16:33 UTC (Tue) by farnz (subscriber, #17727) [Link] (3 responses)

Being extremely pedantic, peak oil means that the amount that you can extract and make a profit on has started to fall.

If a reduction in consumption means that oil prices stay below (say) $50 per barrel, then us knowing that there are 1,000 years worth of oil that costs $60 per barrel to extract is irrelevant to the peak oil concept; it's only the oil that you can extract and sell at a profit that counts.

So, any sane advocate of "peak oil" expects that there will be a significant amount of known oil around after the peak - it's just that it won't be worth extracting, because you'll have to sell it for less than the cost of that extraction.

Or, put another way, we can reach "peak oil" because the replacement of oil with substitutes like solar energy and plant-derived plastics has been so successful that oil is worthless, not just because we've run out of oil

Super long-term kernel support

Posted Mar 22, 2018 5:33 UTC (Thu) by khim (subscriber, #9252) [Link] (2 responses)

If a reduction in consumption means that oil prices stay below (say) $50 per barrel, then us knowing that there are 1,000 years worth of oil that costs $60 per barrel to extract is irrelevant to the peak oil concept; it's only the oil that you can extract and sell at a profit that counts.

Government could always just print more money.

What government couldn't do, however, is to change price of oil measured in terms of oil! Hundred years ago you needed to spend one barrel of oil to extract hundred barrels of oil. Today what you get back is closer to 6 to 8 (it's not easy to calculate the exact "price" since so many components are involved). When you finally hit the bottom (need one barrel of oil to extract one barrel of oil from the Earth) it wouldn't matter how much oil is left there - it's pointless to try to extract it.

Super long-term kernel support

Posted Mar 27, 2018 15:04 UTC (Tue) by nix (subscriber, #2304) [Link] (1 responses)

Well, nearly. It's pointless to try to extract it *for power generation*, but may still be worth extracting using energy from cheaper sources for other purposes (plastics feedstock, chemical engineering etc).

Super long-term kernel support

Posted Mar 27, 2018 15:32 UTC (Tue) by farnz (subscriber, #17727) [Link]

But even using cheaper energy, it's not worth it if the cost of extraction is ¢100/barrel, and the value of oil is ¢50/barrel. It doesn't matter what the value of a ¢ is, or indeed whether governments print more money - if the raw value of the oil is lower than the cost of extracting it, then it will get left alone.

Super long-term kernel support

Posted Mar 20, 2018 17:39 UTC (Tue) by felixfix (subscriber, #242) [Link] (8 responses)

I meant only in the 20-30 year time frame. Obviously oil won't last forever, physically or economically. There may be ways to create petro fuels from ordinary chemicals. But no one can predict that far ahead, and I did not mean to imply that. Sorry for the confusion!

Super long-term kernel support

Posted Mar 20, 2018 18:34 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link] (7 responses)

You can easily create oil from air (carbon dioxide) and water. That's not a problem at all. Just crack water with electrolysis, react hydrogen with carbon dioxide to get methane and then use it to build longer hydrocarbons.

You just need to input a LOT of energy.

Super long-term kernel support

Posted Mar 20, 2018 21:54 UTC (Tue) by rahvin (guest, #16953) [Link] (6 responses)

I agree completely with what you said. But I think the point of the people posting before us was that if trends continue the transportation sector will electrify and the number of uses for black goo out of the ground are going to go down dramatically. As you said all the uses that remain could be satisfied with pure C02 processes or even with plant or other organic inputs along with lots of cheap renewable energy.

Renewables are getting so cheap that we're probably going to see falling electricity prices in the US over the next 20 years and in doing so it's going to price Oil right out of the market. The savings grace for petroleum has always been transportation where there was no replacement for gas/diesel but with the electrification of transportation nearly 80% of oil use goes out the window ending very rapidly the age of oil.

Super long-term kernel support

Posted Mar 21, 2018 15:28 UTC (Wed) by anselm (subscriber, #2796) [Link] (5 responses)

The “age of oil” will stay with us until we figure out how to run long-haul commercial aircraft on something other than hydrocarbon-based aviation fuel (not obvious). Or figure out how to beam stuff (including people) from A to B, whichever happens earlier.

Super long-term kernel support

Posted Mar 21, 2018 16:14 UTC (Wed) by excors (subscriber, #95769) [Link] (1 responses)

That sounds easy - just replace planes with fully-reusable suborbital rockets (like the BFR spaceship, but with hydrogen instead of methane if you want to avoid hydrocarbons). Much more exciting for passengers too.

I've read Glasshouse so I wouldn't trust teleporters not to infect me with malware or accidentally clone me. Especially if the teleporters are running a 20-year-old kernel.

Super long-term kernel support

Posted Mar 21, 2018 16:22 UTC (Wed) by sfeam (subscriber, #2841) [Link]

Never mind "accidentally clone me", what about copyright infringement?

Super long-term kernel support

Posted Mar 21, 2018 16:19 UTC (Wed) by eru (subscriber, #2753) [Link] (2 responses)

how to run long-haul commercial aircraft on something other than hydrocarbon-based aviation fuel

Some carriers are already experimenting with flying on biofuel. Currently it is blended with lots of fossil fuel, but I expect experience and research will eventually allow switching to 100% biofuel.

Super long-term kernel support

Posted Mar 21, 2018 18:29 UTC (Wed) by rahvin (guest, #16953) [Link] (1 responses)

There is also work on electric aircraft engines that aren't propeller driven, don't ask me about them I don't know anything about them just that there are a bunch of companies working to solve it.

Overall aircraft and ship travel combined use about 20% of petroluem, but the ships can easily be converted to electric drive, the trick is aircraft and it's going to be a hard one to fix no doubt. But if Electricity gets as cheap as the projections are showing someone is going to have a LOT of incentive to produce a plane with electric engines with the same performance and speed as Jet Engines.

Super long-term kernel support

Posted Mar 29, 2018 18:39 UTC (Thu) by Wol (subscriber, #4433) [Link]

I gather it's also easy (whether it's fuel efficient or not ...) to manufacture oil. Fill a pressure vessel with plant material, heat, and when it's cooled you have a substance pretty similar to crude oil.

And converting one form of oil to another (especially making petrol and diesel) is pretty easy using zeolyte catalysts, so I think making liquid hydrocarbon fuels from bio-matter isn't a chemistry problem.

It's an economic problem - can we do it efficiently enough to make it worthwhile?

Cheers,
Wol

Super long-term kernel support

Posted Mar 20, 2018 23:14 UTC (Tue) by roc (subscriber, #30627) [Link]

I have a year 2000 model Honda Odyssey with 200K km on the clock. I keep expecting it to fall apart, and we've had a few maintenance items including a transmission overhaul, but generally it seems to run as well as ever. As a software developer, it's humbling.