The existence of LLMs is not the problem. The insane way they've been integrated into society is the problem. Back in 2020, GPT-2 was something novel you'd do random experiments with and something like this would've been cool. As long as it remains a silly experiment, it still is cool.
Every time I see a person leading a model to specialize in a uniquely human endeavour, I think of the last lines of this stanza from Bob Dylan's It's Alright, Ma (I'm Only Bleeding):
Disillusioned words like bullets bark
As human gods aim for their mark
Make everything from toy guns that spark
To flesh-colored Christs that glow in the dark
It’s easy to see without looking too far
That not much is really sacred
The post seems to refer to a very specific subset of demoscene:
> I came across Dwitter.net, which is like Twitter but for 140 character JavaScript demos. They already supported “remix”-ing any dweet so that we can build on top of them.
This person is actually correct. Demos are much more on the art side of coding than most forms of programming, so while LLMs may be up to the mechanical task, they won't produce code with the finesse that a human coder can. That goes double for the art and music. What's more, a demo with AI-gen elements wouldn't be accepted by the community. AI-gen webshit may be embraced by the industry, but the demoscene has different standards: human craftsmanship is the entire point.
Relatedly, I was a bit surprised and delighted to find out that the Sonic (the Hedgehog) community is vehemently rejecting AI recompiles, ports, and fangame projects. Clankers are ok with them when it comes to Dr. Robotnik's evil creations, but not in the real world!
one is close to hundreds of millions in cost and with likeminded stakeholders. the other is hundreds of trillions in costs with the entire human population as stakeholders.
This is quite similar situation indeed. It was once easier to adopt the metric system when the infrastructure was not yet as developed around the previous ones.
Austria-Hungary adopted the metric system in the 1870s. Back then: no cars, no highways, no speeding laws etc. Oxen-drawn wagons in the countryside were not built to a precise spec.
Only the railway, the architects and the artillery had to really switch. Ordinary people would still ask the grocer for half a pound of ham in the 1930s.
USA in the 1970s would have to remeasure all the land communications etc., or end up with a weird hybrid like the UK.
WHat we have in IPv4 and IPv6 is the weird UK-like hybrid...
It's going to take decades, mostly waiting for the non-metric folks to die off probably and the raised-on-metric folks to grow up, but if you don't start you'll never finish.
I'm Australian and the only things measured in imperial units is babies (stones? But I think younger parents now use metric, but my parents only know baby weights in stones), beer (pints, but not the same as overseas I think, also it can vary by state), certain types of land units ("acerage" is a common land plot size) and height (feet and inches, for some reason people don't understand it in metric)
These are all customary usage so I'd say we are completely switched over, we couldn't tell you generally what these units look like outside of "how big is a baby" or "how tall is that person" or "is my beer large or small" really.
AFAIK you all use imperial pints, which are a little bigger than the ones used in countries that still use the customary system (US and I assume Liberia). I'm pretty sure your pints are the same as those used in the rest of the commonwealth, but corrections are welcome.
That goes back to the fact that the imperial system replaced the older customary system after the American revolution. The US legislated a somewhat simplified version of the Customary System the English were using at the time, which was a piecemeal system based on several different acts of parliament over hundreds of years. We (much) later made slight changes so that our measurements are based on metric.
The Imperial System was instituted in the UK in 1826 and most of the measurements stayed the same, which is why our foot and pound are the same as your foot and pound. The biggest change was that imperial simplified the various liquid measurements to make them uniform. Before then, different liquids used different measurements. The gallon used in the US was originally based on the wine gallon whose value had been last set by parliament in 1709.
Canada also changed over in the 1970s. I was told as a kid that the US had planned to switch with us but backed out at the last minute. I'm not sure if that's true or not. Today, the imperial system holds on in a few places, like heights and weights of people (although not at the hospital) and cooking temperatures.
IPv4 is better than IPv6, and imperial is better than metric. I would like something to replace IPv4, but that thing shouldn't be IPv6 and probably won't be at this rate.
IPv4 is inherently more user-friendly. IPv6 was designed for a world, in the 1990s, where it was expected that every host wanted to be routable. SLAAC etc were all natural consequences of that. For today's use cases, the v4 NAT public/private thing actually turned out to be nicer than v6 global routing. There's still confusion today about whether devices are exposed on a v6 network because the spec itself says default-deny firewall is optional. They had to recently add privacy addresses. The other kinds of v6 addrs like link-local and ULA add more complexity. V4 addresses are easy to remember, which v6 advocates keep saying doesn't matter but it does. There are all these little network management details that are messier in v6, as you will find on online help forums for v6 advocates who you cannot write off as just being too used to v4.
The v6 spec also has all these alternative ways to use it more like v4, commonly brought up as a cop-out against the above criticisms, but that's actually part of the problem. The spec is too big, and things only implement a subset of it, see Android not supporting DHCP6. Realistically you'll have 99% of people using anything like this the default way, and the spec does define SLAAC, no-NAT, etc as default.
But you could say v6 is better just cause it has more addresses, and that's fair.
...Imperial is better than metric because it lines up with everyday usage better unless you're specifically concerned about the physical properties of water. Especially ˚F vs ˚C. Not a big deal though.
0-100˚F lines up well with weather ranges, and 1˚F is about the smallest diff someone can feel. Many digital thermostats use integer ˚F and introduce 0.5 steps if you switch to ˚C, even in countries where ˚C is default, so they kinda admit it.
Pounds vs kg, similar story as F vs C. Pound is pretty much the human-discernible granularity. Otherwise doesn't matter.
0-100 MPH lines up well with car speeds, which is usually the thing you're measuring with that. KPH is excessively granular for that. Probably not by design, but yeah.
The feet/inches vs meters/cm thing matters less. It's base-12 vs base-10. You can do decimal feet or fractional meters of course, but it's less common, so apparently both ways are fine.
As a Canadian I have no idea what 72˚F means or feels like: is it warm or cold? I do know what 22˚C feels like and how to dress for it. How "intuitive" a temperature number is completely learned behaviour and cultural acclimation.
Pounds vs kg: see above. All the non-American people seem to have "intuitive" grasp of metric measurements.
> 0-100 MPH lines up well with car speeds, which is usually the thing you're measuring with that. KPH is excessively granular for that. Probably not by design, but yeah.
So 'more granularity' is good for temperature but not for speed. Got it.
> The feet/inches vs meters/cm thing matters less. It's base-12 vs base-10. You can do decimal feet or fractional meters of course, but it's less common, so apparently both ways are fine.
By "foot" do you mean 'regular' foot or survey foot:
Because there are two different units in the US customary system, used for basically the same purpose, that use the same name. And the "foot" isn't the only unit like that:
Here's another one: which weighs more, a pound of feathers or a pound of gold? A pound of feathers. Because feathers use one 'pound-system' [1] and is 453g, while gold uses another pound-system [2] and is 373g. But of course an "ounce" of gold (31g) is more than an "ounce" of feathers (28g), because Troy is 1/12 of pound but Avoirdupois is 1/16. In addition to the above "ounces" there is also "fluid ounces".
I didn't say it's about being used to one. It can be "in the 20s" C for multiple months, kinda silly. 22.0˚C also feels different from 22.5˚C. There's a right amount of granularity for temperature and for speed.
Foot and pound have unambiguous meanings in the US. There's a reason the whole survey foot and gold-pound thing is trivia that nobody knows, they're niche or abandoned systems that you only use if you're being clear that it's not regular feet or pounds. I also know that the US system is technically not "imperial," but nobody cares.
I'm not the person you were replying to, but I can answer this one. And yeah, I can see all the metric fanboys (a very small subset of metric users, mind you) hovering over that downvote button, but I don't get paid in Internet points so here goes:
The nice thing about imperial (as well as the customary system used in the US) is that it's based on measurements that people have found useful over thousands of years. They're human units, adapted to what people need to do with them, not the other way around.
It's difficult to explain to metric folks because it's really something you have to experience. But here's an example:
Reservoir volume in the US is measured in acre-inches. An acre is a unit of area used for measuring property, usually farmland. It's a bit less than half of a hectare. And of course an inch is 2.4cm. Acre-inches sounds totally bizarre when you first hear about it. But for the people that need to work with this value, it's an extremely useful way to measure reservoirs and gives you an intuition that you just don't get from cubic meters.
It's true that you can't just convert that to other units of volume without doing some math or looking up the conversion rate, but that rarely matters. If you're trying to figure out how fast the lake will rise or fall with a given inflow and outflow, you're going to have to have some way to account for the fact that the lake is a) an irregular shape and b) takes up more area the higher the water gets. You already need to know the area of the lake at any particular height to work out the evaporation rate anyway, so that's a number you're going to have. The units are designed to make that calculation easier, which was really important back in the days when slide rules were the state of the art.
All the customary units are like this in one way or another. They're units designed for a particular domain, and when used in their domain they work very well and calculations can be done in your head. They use fractions because that's what people have historically used (seriously - my grandfather could work fractions in his head faster than I could type them into a calculator, and he made asphalt for a living. Imagine what a carpenter could do.).
Think about it. Imagine you were never taught a measurement system. You find yourself needing to come up with a unit to make measurements for wooden planks you're going to use to build a house. Would you choose a unit as long as a meter? Or one as short as a centimeter or even a decimeter?
Bonus challenge to the fine folks clicking that downvote button: give me a real practical reason to prefer Celsius to Fahrenheit, besides the fact that Celsius is part of the metric system. No, the freezing and boiling points of water don't cut it - I work with hydrocarbon fuels, I salt my driveway and patio, I don't live at sea level, and quite honestly every American 8-year-old can spout off 32° and 212°. There's a Fahrenheit equivalent to Kelvin (Rankine) so the Kelvin to Celsius equivalence doesn't count either. My claim is that both scales are completely arbitrary, and only advantage one has over the other is that you're used to one of them. Go ahead, prove me wrong!
> Reservoir volume in the US is measured in acre-inches.
Never heard of an acre-inch before, and I'd have to get out a calculator to work out how much water that actually is in pints or gallons, or how much it weighs.
The metric equivalent would be a hectare-meter or a hectare-mm. I know a hectare is 100 m x 100 m, so 10,000 m². So I instantly know that a hectare-meter is 10,000 m³ (10 ML, or 10,000 tonnes of fresh water). A hectare-mm is 10 m³ (10 kL). If my 1 ha block gets 1 mm of rain, then that's 10 kL or 10 tonnes of water. So easy i can do it in my head.
I only know of it because it's a unit used by the US Army Corps of Engineers, who maintain the local reservoir. A buddy of mine's dad worked for them when I was a kid so that lake was my stomping ground.
A couple more to add to your list: an inch is about as long as your thumb joint (with some variation between individuals obviously). It's an easy way to gauge the size of everyday objects without needing to carry around a tape measure.
The recommendation for emergency supplies is about a gallon of water per person per day - or 3.7 liters if you're using metric.
I've also heard it claimed that the 0-100 Fahrenheit scale is more functionally useful because the endpoints of 0 and 100 represent the tipping point of serious danger for humans from hot or cold weather. The closer you get to 100 degrees the more care you need to take about heat stroke and keeping hydrated, and similarly the closer you get to 0 degrees the more care you need to take about retaining heat and staying out of the elements - the freezing and boiling point of water isn't a necessary marker for survival. And people like nice round numbers to serve as markers for such things, divisibility be damned.
Metric is sterile and functional for calculation and consistency, which is fine and necessary for many things, but that's not a reason to completely ditch Imperial.
> The nice thing about imperial (as well as the customary system used in the US) is that it's based on measurements that people have found useful over thousands of years. They're human units, adapted to what people need to do with them, not the other way around.
US customary is not "thousands" of years; temperature (Fahrenheit) only came out in 1724 for example, and many other units were only standardize much more recently:
Second, even with-in the system itself there was also sorts of inconsistencies. Until just a few years ago there were two "foot" units; one was recently retired:
Here's another one: which weighs more, a pound of feathers or a pound of gold? A pound of feathers. Because feathers use one 'pound-system' [1] and is 453g, while gold uses another pound-system [2] and is 373g. But of course an "ounce" of gold (31g) is more than an "ounce" of feathers (28g), because Troy is 1/12 of pound but Avoirdupois is 1/16. In addition to the above "ounces" there is also "fluid ounces". And let's not forget about the third system [3], which got replaced by… the metric system in 1971.[4]
> Think about it. Imagine you were never taught a measurement system. You find yourself needing to come up with a unit to make measurements for wooden planks you're going to use to build a house. Would you choose a unit as long as a meter? Or one as short as a centimeter or even a decimeter?
Millimeters. That single unit gets you down to very small scales (finer than 1/16") and to large lengths: four significant digits (9999mm) gets you to the equivalent of >32' (~11yd).
You misunderstood me. It's based on measurements that people have been using for thousands of years. The Romans had feet and miles, for example - not exactly the same lengths, but for the same purposes. The Japanese shaku is almost exactly the same as a US foot. A Japanese Shō is about half a gallon. A Chinese jīn is just slightly larger than a pound. There's a reason for this.
> there was also sorts of inconsistencies
I never claimed otherwise. But do notice that the survey foot and standard foot weren't used for the same purposes, and the difference between them wasn't likely to confuse anyone except maybe people new to the field.
> Here's another one: which weighs more, a pound of feathers or a pound of gold?
Yep, Troy weights. And no one's sure where the Troy system came from. The main reason for its use is because of international trade. Also not something regular people (i.e. people that didn't work directly with precious metals) would have had contact with.
> Millimeters
Hah! Yeah right. Look at all the various non-metric systems that have existed throughout history. Did any of them use something like millimeters to measure wooden planks? Why not?
> As evidence for this Australia/NZ/EU do everything in millimetres in construction.
They do it because the metric system is mandated there and it doesn't give you reasonably sized unit for that scale of work.
Try this: go to a home center (or whatever it's called there - the place you buy building supplies) and measure some drywall. Do the measurements you made agree to the advertised ones within half a millimeter? Compare the dimensions of drywall sheets from different pallets and manufacturers. Do their dimensions agree to within half a millimeter? If not, that's a sure sign you're using units poorly adapted for purpose.
IPv6 has a huge amount of bloat created by guys in cargo shorts and pony tails being flown out to exotic locales 3x a year for IETF conferences to argue about minutia and try to get their name on some IPv6 spec document. All they needed to do was add a couple more dots onto IPv4 and maybe switch to hex.
> All they needed to do was add a couple more dots onto IPv4 and maybe switch to hex.
There is no 'just' extending the address length:
So it's a matter of mathematical and physical fact that to expand
the address size, you must change the protocol, and that means
[three] things immediately:
You have to change the version number.
You have to add new code to handle the new version.
[…]
You need a coexistence technique so that updated systems,
with the new protocol, can connect to old systems that know
nothing of the new protocol.
Two minutes of thought show that this third requirement has only
two solutions:
(3A) Dual stack, in which the new machines speak both
the old (IPv4) and new (IPng) protocol.
(3B) Translation, in which something translates addresses
between the old and new protocols.
You also have to update DNS to create new resource record types ("A" is hard-coded to 32-bits) to support the new longer addresses, and have all user-land code start asking for, using, and understanding the new record replies. Just like with IPv6. (A lot of legacy code did not have room in data structures for multiple reply types: sure you'd get the "A" but unless you updated the code to get the "A+" address (for "IPv4+" addresses) you could never get to the longer with address… just like IPv6 needed code updates to recognize AAAA, otherwise you were A-only.)
In any 'address extension' plan the legacy code cannot use the new address space; you have to:
* update the IP stack (like with IPv6)
* tell applications about new DNS records (like IPv6)
* set up translation layers for legacy-only code that cannot be update to reach extended-address-only destination (like IPv6 with DNS64/NAT64, CLAT, etc)
Wasn't it from academics in their ivory towers? That's kind of the opposite of flying out to exotic locales.
Of course, back here in the real world you can look at the RFCs yourself and see that they're credited to places like Ipsilon Networks, Xerox and Sun Microsystems. Boring corporate network companies.
No literally, the IETF meets 3x a year in a different location each time and all the guys from with cargo shorts from the companies you named argue about IPv6 minutia all day and spend their per diem by night. It's actually a lot of fun.
Some of those locations probably did feel more exotic back in the early 90s than they do today... but they're also sensible places to hold international conferences. Where are you supposed to do them, if not there?
IPv6 invalidated the existing IPv4 addresses, changed the addressing scheme, and de facto got rid of DHCP and NAT in favor of SLAAC and global routing. It doesn't share any state with v4. This was a very different set of choices from just adding more bits to v4.
Nobody's stopping you from using DHCP and NAT on IPv6, it's just not the norm.
The protocol itself is literally very similar to IPv4. They got rid of the IP header's checksum, but I don't think that's something you're complaining about, right? They renamed Time-To-Live to Hop Limit because that's what it actually does - is that a complaint you have?
What do you mean it doesn't share any state with v4?
By not sharing state, I mean v6 doesn't honor v4 routes/addresses, DHCP state, and DNS entries. They made all-new address blocks. This seemingly never gets talked about for how big of a deal it is. That makes it so enabling v6 instantly means reworking your entire network, not just speaking a different protocol.
Say I have 1.2.3.4/32 in v4. Let me keep that in v6. Just getting everyone onto v6 is a big enough change without also changing the routes. If I talk to 9.8.7.6 and it's v4-only, fine, the last v6 hop translates to v4. Even DNS doesn't strictly need to change yet. Then let's say enough people get on v6, so I hand out 1.2.3.4.5 to some host. If 9.8.7.6.5 wants to send to that, but there are some v4-only routers in the middle, no problem. The last v6 router translates to v4 dst=1.2.3.4 and puts the extra address bits in the payload, then when it reaches my router, it knows to rewrite as 1.2.3.4.5. Even in this later state, you really only need v6 support on the ends and any routers going deeper than /32.
DHCP and NAT on LAN could share state too. A host at 192.168.1.2 should be able to just turn on v6, it's still 192.168.1.2, some other host gets leased .3 instead, and the router knows to translate if .2 and .3 want to talk. I don't want it to suddenly be FC00::whatever.
I have no complaint about the v6 header, and a subset of the v6 protocol is fine because it's indeed just v4 w/ extra bits like you said. That ipv5/whatever proposal people keep making is doable with most routers/hosts just seeing it as ipv6 w/ DHCP+NAT.
And yeah, today I could use DHCP6 if I really wanted, but it'd mean lots of incompatible devices get either booted off my network or put on a separate SLAAC one, so it's not a serious choice. NAT66 is way more doable but also rare on consumer routers. So my only complaint about the protocol is that the defaults are backwards.
Nah, the details in the spec all have good reasons, but it's based on the original mistake of making it an entirely separate network from v4 with new address blocks.
> […] but it's based on the original mistake of making it an entirely separate network from v4 with new address blocks.
This was not a "mistake":
So it's a matter of mathematical and physical fact that to expand
the address size, you must change the protocol, and that means
[three] things immediately:
You have to change the version number.
You have to add new code to handle the new version.
[…]
You need a coexistence technique so that updated systems,
with the new protocol, can connect to old systems that know
nothing of the new protocol.
Two minutes of thought show that this third requirement has only
two solutions:
(3A) Dual stack, in which the new machines speak both
the old (IPv4) and new (IPng) protocol.
(3B) Translation, in which something translates addresses
between the old and new protocols.
> Yes it's 3A. They didn't need to make the new protocol have separate state from the old one.
Except there will be a period of time when many 'network elements' (routers, firewalls, servers (web, SMTP, DNS, etc), clients) will not be updated for the new protocol so you may also need 3B.
And this will be true for any IPng, and not just the one that just happened to be chosen to be for "IPv6" (which was SIPP, over CATNIP and TUBA; see RFC 1752).
> Let me keep my old address on the new protocol.
You can/could. You can encourage people who have addresses in the older protocol to use the new one with "free" "automatic" address assignments in the new one to encourage translation and tunnelling:
> For any 32-bit global IPv4 address that is assigned to a host, a 48-bit 6to4 IPv6 prefix can be constructed for use by that host (and if applicable the network behind it) by appending the IPv4 address to 2002::/16.
> For example, the global IPv4 address 192.0.2.4 has the corresponding 6to4 prefix 2002:c000:0204::/48. This gives a prefix length of 48 bits, which leaves room for a 16-bit subnet field and 64 bit host addresses within the subnets.
And the 'old' IPv4 addresses are available in the new IPv6 world:
> Addresses in this group consist of an 80-bit prefix of zeros, the next 16 bits are ones, and the remaining, least-significant 32 bits contain the IPv4 address. For example, ::ffff:192.0.2.128 represents the IPv4 address 192.0.2.128. A previous format, called "IPv4-compatible IPv6 address", was ::192.0.2.128; however, this method is deprecated.[5]
6to4 and NAT64 are not what I'm saying. Those involve a relay server and v4 translation, and there's an entire memo on why that's unreliable and abandoned. I want to switch to fully native IPv6 without changing any of my addresses. There's no reason this should be impossible.
> I want to switch to fully native IPv6 without changing any of my addresses. There's no reason this should be impossible.
… except for the fact that you have to have completely new data structures and packet format to fit more address bits, and once you change the structures you need a new version field so any gear that sees a packet knows where to use the old code paths or new ones.
Then you're at the point where you have to push new code everywhere to handle these changes, and some nodes will have only the old code and some the new code, so now you're into translation and/or tunnelling to make sure the old addresses can talk to new addresses. Depending on their risk tolerance, some operators will enable the new code and some will not.
You also have to update DNS with new records, as A records are hard-coded to 32 bits. So now you're updating DNS libraries and user land.
There is no way to expand address space without going through all the same issues as what we call "IPv6" (originally SIPP) has gone through.
I know. All these things were possible and completed for IPv6, but that's only part of the battle. The rest is making it easy for users at various levels to migrate. IPv6 didn't do that, that's why we're stuck in today's situation.
IPv6 added an unnecessary requirement to change all existing routes, which is huge. If still using old v4 addrs in v6, there should be no need to change DNS yet. You shouldn't need a different DNS or DHCP server, only one that supports both protocols (also shouldn't be forced to use SLAAC). V6 translation to/from v4 should be doable at any hop instead of relying on a flaky anycast relay. The later stage of migration should just be that people start using longer addresses given that both ends are v6 and DNS etc supports it then. Even still there should be no requirement for every hop along the route to be v6, only the ones that go deeper than /32 near the ends. But we got all these extra constraints because of ipv6 design decisions.
Why spend all this energy complaining about v6 not doing something that it did do?
> The later stage of migration should just be that people start using longer addresses given that both ends are v6 and DNS etc supports it then
It seems your only real problem is that you completely missed the early part of the transition. We've been in what you're calling the later stage for twenty or so years now.
IPv4-compatible v6 address was only for dual-stack hosts talking over a v4 network. They still had their own v6-native addresses that they had to use if they wanted a v6 network. So this isn't what I described where the v6-native route is the same as the v4 route, or a superset in the later stages (shared down to /32, v6-only in deeper subnets).
The thing I want to solve isn't just v4 routers in the middle, that's more of a bonus. All the routers between my machine and any service probably support v6 anyway, but a bunch of services are v4-only because of the migration effort. They should be able to turn on v6 without having to change everything about their networking.
Wasn't "dual-stack hosts talking over a v4 network" what you wanted?
v4 routers don't understand and can't route v6 packets, so you can't use the same route for both of them. v6 packets have to go to a router that understands v6, and v4 packets have to go to a router that understands v4.
I have no idea what "or a superset in the later stages (shared down to /32, v6-only in deeper subnets)." is trying to say.
It's perfectly possible to turn v6 on without changing anything at all about your networking. When e.g. Windows Vista came out, hundreds of millions of people did exactly that.
Why can't the service just set up NAT64 to accept your v6 connections on their v4 servers?
I wanted switching to v6 to involve the minimum necessary changes. You keep your old /32. If you're sending from a /32 to a /32, it follows the same path of routers that an ipv4 packet would follow, that's what I meant by same route. There could be a v4-only router in the middle requiring translation, but I'm not even assuming that's a common issue at this point.
"or a superset in the later stages (shared down to /32, v6-only in deeper subnets)": The ipv6 state would be a superset of the ipv4 state. So if a router owns 1.2.3.4 in v4, it also owns that in v6. The ownership of 1.2.3.4.5 below that would only be part of v6 state.
What we got instead was two separate states, or parallel networks. Switching to v6 means changing everything all at once.
> Why can't the service just set up NAT64 to accept your v6 connections on their v4 servers?
It's not that simple, otherwise they would've done it. They have stuff like multiple addresses, geo DNS, and rate-limiting. Separate issue, AWS was v4-only for a long time and still treats it as more first-class because of how many internal things need to change for v6, well beyond just having hardware + OSes support it.
They don't need to be really long. Adding one more octet to v4 makes it 1 trillion addresses. Sure the header should support longer than that, but it doesn't mean everyone needs to use all of it.
1 trillion addresses isn't anywhere near enough[1]. Remember that we're not talking about L2 here. (...I don't think 1 trillion MACs would even be enough if we want to ensure they're unique, and it seems the IEEE agree with me since they expanded MACs to 64 bits, but that's a different conversation.)
And you already don't have to use all 128 bits in v6. You're forced to use about 48 of them, but the rest is your choice. You can just... not use some of them. It's fine.
The world is already managing with 32 bits of addresses plus NAT, and NAT most likely gives you less than 256 concurrent users per address. But sure, if some cheap ISP only owns a few /32s and needs to use more octets, let them do that, doesn't mean my 4-octet address needs to be super long too.
You're kinda forced to use 128 bits in v6 if you're an end user, cause that's the lowest common denominator. SLAAC is default, DHCP6 is widely unsupported. Most ISPs also only give out a /64 per spec minimum.
Will you repeat the whole dance when you need a second extra octet? There are good reasons IPv6 is designed the way it is.
They considered 64 bit addresses at first but they didn't want to ever risk running out again so they doubled it again.
Try to remember it's not just about the sheer number of addresses - you also want to allocate them in big blocks so that few people ever need to expand their blocks, and they can split their big block into more blocks that are still big.
No, the field should be 128-bit to have headroom, but you don't need to hand out /128s to everyone. Everyone with a /32 now would be fine with /32 or /40.
> IPv4 is perfectly and objectively fine on internal networks. IPv6 is overkill and way too complicated for most.
Unless there's a merger/acquisition and the two companies both use 10.0.0.0.
Most home networks use 192.168, and so you can't use that in corporate environments, because the WFH people who VPN in could then have routing snags between local and remote locations.
Also, in your VPN software be sure to perhaps not use 172 addresses (like my previous job did), as some of those ranges are used by Docker by default, and so you can cause issues for developers.
But other than that IPv4 RFC 1918 is not complicated at all…
IPv6 discussions keep having this scenario - which is barely relevant to how nearly all IPv4 LANs operate.
It's assuming a huge zero-trust point-to-point network where it's a rare scenario for IPv4 LANs. IPv4 LANs tend to assume the internal network is semi-trusted and route through hubs, so merging networks is a bad idea even when the mapping does not conflict (which often doesn't if the companies are tiny like most are). Even in a zero-trust point-to-point environment you'd want to review the other side, convert it to your software, etc. And the people on the being acquired side have little reason to care too.
It's a scenario that's only relevant for a huge company constantly acquiring other companies that operates via zero-trust point-to-point networking and strongly tries to minimize human review - basically Google/Alphabet.
In theory IPv6 solves the problem by putting all of this under a single prefix (fc00::/8) and asking nicely that implementation/admins add 40 random bits to form their prefix. Which is plenty to prevent (almost all) eventual conflicts.
The thing is, asking nicely doesn't always work. Sure, automated tools like docker/incus/wireguard/tailscale will respect it. But admins who don't want to be stuck typing long addresses? Not so much. I've lost count of how often I encounter a network admin that manually set their prefix to some fcXX::, when it's not outright fc00::.
Of course you can (and should) blame this on admins and not on IPv6, but then the same could be said of IPv4 where adding entropy to 10.X.X.0 is the usual recommended way of doing things.
> In theory IPv6 solves the problem by putting all of this under a single prefix (fc00::/8) and asking nicely that implementation/admins add 40 random bits to form their prefix. Which is plenty to prevent (almost all) eventual conflicts.
IPv6 solves this by making it easy for everyone to get their own 200::/3 address so that when two companies get together they have completely different /48s (or whatever).
> […] but then the same could be said of IPv4 where adding entropy to 10.X.X.0 is the usual recommended way of doing things.
There has never been a recommendation to add entropy to 10/8 addresses. At least I have never heard of such a thing. Can you provide a citation for this (alleged) recommendation? Is there an RFC of some kind?
I have seen people suggesting this before, but it's problematic for a few reasons:
- If you are managing the 4to6 mappings manually, it's obviously not going to be scalable. Just to give people an idea, websites nowadays can connect to 10+ IPs in a single page due to CDN load balancing and stuff.
- If you are handling the mappings automatically, such as by monitoring DNS queries and creating dynamic mappings based on that, the issues become more insidious: applications with hardcoded IP literals would fail without an obvious reason.
I have encountered multiple IPv6-only domains - certain random domains were accessible through Chrome, but unreachable from any command line tools. Turns out they were IPv6-only, and Chrome was doing its own DNS resolution, while anything that was using the platform DNS stack was hitting a black hole between OPNsense and my ONT.
It's the hobbyists who LARP as network engineers, and that end users and hiring managers can't tell the difference, who are holding everyone back.
Anyone who doesn't by default set up IPv6-only networks by default with NAT64 and DNS64, or who present any notion that IPv4 is somehow "okay", shouldn't have a job anywhere, not even an entry-level desktop support job.
If you have a dedicated IPv4 address, those behind NAT and CGNAT face significant throttling and false positive blocking. It's been made worse by scraper botnets.
Creating an IPng was always going to be complicated:
So it's a matter of mathematical and physical fact that to expand
the address size, you must change the protocol, and that means
[three] things immediately:
You have to change the version number.
You have to add new code to handle the new version.
[…]
You need a coexistence technique so that updated systems,
with the new protocol, can connect to old systems that know
nothing of the new protocol.
Two minutes of thought show that this third requirement has only
two solutions:
(3A) Dual stack, in which the new machines speak both
the old (IPv4) and new (IPng) protocol.
(3B) Translation, in which something translates addresses
between the old and new protocols.
This is very valuable, but not sure if you think it contradicts my comment.
I'm not saying that IPv6 is incompetent, or that IPv6 implementations break backward compatibility, or that the problem can be solved without breaking bc.
I'm saying that while IPv6 breaks bc itself, dual stack does not, and even full migrations can maintain bc or gracefully degraded.
In essence, not all implementations IPv6 are good, and not all migrations from IPv4 are responsible, if someone is blindly team IPv6 or blindly team IPv4 in general there's a lack of nuance.
It's impossible for a long address system to be compatible with a short address system. No one has ever found a way to make it work. Plenty of people think they have, but upon closer inspection, they haven't.
NAT is the only one that actually does work, and it's terrible.
In another decade this headline would be "please let books die" or "please let tv die". Some people are biologically wired to process information in a maladaptive way (so much research about this), and the medium is irrelevant.
I say this that even simple text based sites like hacker news can be used inappropriately.
CD ROM encyclopedias were only really viable when dial up speeds limited the amount of multimedia content on the internet. Both Microsoft and Britannica were victims of internet encyclopedias. And now static internet based encyclopedias are themselves under threat of personalised AI content as well. The pursuit of knowledge keeps changing what knowledge itself looks like.
Britannica actually had a great website during the dot com era - hyperlinked, very browsable, and definitely had that semi-addictive "rabbithole" quality.
The hurt themselves badly by imposing a really annoying paywall just as Wikipedia was beginning to take off. A hybrid model with tiered access (maybe even a limited ad-supported free tier) might well have kept them on top - instead, they offered only a single expensive annual plan priced to compete with the paper or CD editions.
Backwards compatibility has limits. When a piece of software has been abandoned for 30 years, there is very little commercial value in supporting it. Apple, Google and even Linux are more aggressive in dropping support. It's only IBM on mainframes with multi million contracts that is really going for backwards compatibility at any cost.
Doing something that engages your brain in a unique way rewires your brain. I'm currently learning Astronomy and it is changing my thinking already. There are so many academic fields out there that each wires your brain in a unique way. We say were are in the age of AI but human intelligence needs to be studied even more than ever.
I can't even run a binary from one Linux distro to the other, let alone one from a random Unix from the 80s. If a rebuild from source is allowed (which would need to allow modifications to the build system) that's a completely different game.
I had to run a Linux binary compiled for Ubuntu 10.something a few years ago. I couldn't get the drivers to work properly in libvirt and getting the binary to run on modern Linux was practically impossible, even with 32 bit support installed.
The downside of dynamically linked binaries: every tool of medium complexity needs a close-enough version of a common library, and once those get upgraded, the tool dies.
Most statically compiled version should work with a compatible enough libc, but I haven't seen many old programs like that.
Wine, on the other hand, runs tools from the 90s just fine, as long as they're 32 bit.
You don't need to go full static linking. Just bundle every library that doesn't have string backwards guarantees or rather bundle everything except base system libraires. The same is also how things work on Windows (and thus Wine really). Wine won't be able to run a binary that links to some random libpng or whatever that the compiler found in your System32. The only difference is that Microsoft makes an SDK that gives your some idea of whats a system library you can rely on as well as the ability to target older versions than the system you are building on. The latter can be done on Linux as well but there is no-pre made SDK so the easiest solution ends up being a VM or chroot with older libraries.
> Just bundle [...] everything except base system libraires
That means everything above libc. There are no "base system libraries" in the Linux world for GUI. Not even libX11, that has been deprecated in favour of libxcb, which has been deprecated in favour of not having X11 and running Wayland. Some days I am amazed that the ELF binary format hasn't been deprecated yet...
LibX11 is still alive and kicking. It is however possible to bundle libX11 and/or libxcb (libX11 today sits on top of libxcb) because the wire protocol between them and the X server is also backwards compatible.
And of course Wayland has XWayland to continue compatibility. This doesn't give you access to new toys but neither does restricting yourself to a 30-year old Win32 API.
And beyond libc/libX11/libxcb, libasound has also provided good backwards compatibility with newer systems implementing shims to make it work. You can also use libpulse directly if you don't care about old/opinionated systems though.
Many programs are dynamically linked to at the very least, libc. If the system libc is older than what the binary was developed for, it will often fail. Failures going the other direction are less common, but can and do still happen. If the binary was statically linked with everything it needs, you've got a pretty good chance it'll work, but that's a bit harder to do with GUI programs that need to interact with X or Wayland.
It depends. I've found for example, if i compile on Ubuntu or Debian, it will probably work on any Ubuntu, Debian, or Debian-based distro released at or newer than the one I compiled it on. There are other cross-compatible examples.
More broadly, there are things like manylinux and MUSL which allow cross-distro applications.
I agree - the model I use for why containers (Docker etc) exist is to abstract over Linux ABI incompatibility. I don't think they're a great solution, as it complicates software distribution. I should be able to distribute an executable and have it just work.
Linux isn't the source of the incompatibility. The kernel itself is very strict about not breaking userspace. Your userspace is breaking compatibility on its own, containers fix it by shipping the userspace components within the container, which only works because the kernel maintained compatibility. If the kernel broke compatibility you would need to use a VM so you can run a different kernel version.
Hah well yes for Windows. This is a mercy, but my mindset is still "Works on Windows, Linux and Mac". (The last being dropped for anything that requires Cuda...)
From the article, the time to beat is November 16, 1992. This might be possible on one of the BSDs, as FreeBSD and NetBSD both document support for running a.out executables with the 4.3BSD ABI. I guess it would depend on whether anyone had ported X11 to run on a BSD on i386 before then. Alternatively, maybe you'd be able to get something running on SCO OpenServer (if you count that as a current OS, it IS still getting security patches). It documents support for XENIX x.out binaries ( https://www.scosales.com/ta/kb/126704.html ) so you might be able to run software compiled in the late 1980s for XENIX/386.
NetBSD has excellent binary compatible with 80s "unix", the problem is most (all?) of the 80s unix platforms it supports aren't x86.
Current NetBSD still has good support for running SunOS binaries, though, including graphical ones. I'm certain you could get a SunOS 4.x binary from pre-1992 to run on NetBSD/SPARC. I've no idea how far back you could potentially go on NetBSD/m68k -- probably quite a ways, especially if the binaries are statically linked. SunOS pre-1988 was all static linking!
(Running "real" 4.3BSD binaries might be hard -- I think NetBSD/VAX defaults to ULTRIX compat, which is an either/or thing versus strict Berkeley Unix compatibility. Same binary format, similar but incompatible syscalls/data structures.)
Random people that you're expecting to look stuff up for you aren't customer support; they're comparable to 404 Media, except they don't even make a living from views. They're simply random people that you're taking advantage of.
People linked you to lmgtfy because they were saying you were lazy or stupid, not because they weren't going to support their products. They hadn't sold you anything.