Ethernet connector wiring is so infuriatingly unnecessarily frustrating. It's fiddly, it requires too much micromotorics, it's error-prone, and if you're colorblind or work in dark spaces, you're fucked.
I wonder, why do we even HAVE to bother with wiring order at this point? Shouldn't we really have some kind of smart wire order autonegotiation algorithm by now? Like, whatever, use any color order you like, including flat wires, it really shouldn't matter which order your connector is wired in, as long as all conductors are set, it shall work.
@vovanium
You don't have to. They're all twisted the same from the factory, and as long as you transmit corresponding signal through corresponding wire, it uses the same twisted pairs.
@drq Signal is transmitted through a pair, not a single wire. If you transmit, say +1.5V signal through wire 1, you have to transmit -1.5V to a wire it twisted with. If you assemble the cable randomly, you have to detect every pair.
not only proper pin-to-pin connection is important for ethernet :-) proper physical wire too, because electrical interference between wires can disrupt signals on high speed/long distance. so you need also an reflectometer or something like that per ethernet port :-)
@drq they are twisted and twisted certain way and certain pair must be connected to the certain jack pin. if you connect pairs to a jack random way, it will work occasionally. np, just let’s call it “casual ethernet” it will be funny.
@drq I once saw a bank lan 10basetx, built using untwisted flat cable. it worked. with crc errors, packet loses etc, but worked against all odds. why not :-)
@drq ethernet - это гениальный интерфейс, я удивляюсь, зачем нужен USB, когда есть ethernet!
4 провода, скорость 10 или 100 Мбит + питание до 100 Вт, аж на 100 метров. + ещё 4 провода и экранирование, и доступны стандарты 1 и 10 Гбит. Коннектор не удобный (толстый), можно другой придумать(. При обжиме можно перепутать провода в паре или пары между собой, и оно продолжит работать (главное между парами провода не менять и не включать в такой кабель PoE). И куча уже готовых протоколов для всего разного, зачем нам USB?
@drq we are so not going to like how costly and unreliable that'll make a lot of the networking hardware. So many additional moving parts and failure states, most of them utterly unfixable in the field. And considering PoE exists, every single one of those smart conductors will have to be quite beefy too.
But we could at least color the wires in a way that's easy to piece together intuitively and without relying on color hues. Green lightness gradient? High-contrast dash pattern of varying lengths?
@drq memorise and then? Like, imagine there is signal at physical wire №2. But because order is random, wire №2 may correspond to... any «logical» wire. So you have to go where you memorised wire order, and see, that №2 is actually №8. Repeat on every signal.
@mo and then transmit the data according to the order you've negotiated, until the plug is pulled and connection is lost. You don't have to repeat that on "every" signal. The order doesn't change randomly as long as the wire is connected.
I mean, hell, we already do negotiate between A and B wiring standards where four wires are switched. It's not much different.
@mo@drq But order really could be determined in parallel. I guess it could be done even on hardware level. Like we define distinct signal forms for each input. Upon connection the device assumes "classical" scheme and awaits input. If no valid input, the negotiation cycle starts. Our device reads all inputs for incoming "signature" signal. If received, inputs are mapped internally and the existing Ethernet protocol module is connected to the wires. If still no input, our device at random interval broadcast signatures(assuming classical pins wiring)/listens to the answer until valid Ethernet packet arrives (that means our schema was accepted) or signature arrives (then we accept schema and send back Ethernet packet)
@mumzik I can do you one simpler. Make no assumptions on wiring order in the first place. Upon connection, burst a rainbow of frequencies on the conductors 1 through 8. There you go, what comes out the other side, is your conductor order for the day. Write that down to network device's RAM (the same RAM you save ARP tables to). You will roughly need 1 byte for that (8 wires make 256 combinations). Send "ready" signal back, repeat that on the other side, and you're golden.
@mumzik@drq@mo I'm showing that dealing with high speed transmission lines is not that easy as you might expect.
The model shows only two aspects of this: crosstalk and lack of common grounding.
There are also other issues you must deal with.
DC decoupling. Normally ethernet ports are decoupled by transformers, ignoring this you'll have to deal with high DC voltage offsets (tens or even hundreds of volts) and high common mode noise.
Proper waveguide parameters of the board traces. Normally each twisted pair continue on a PCB as differential microstrip line matched in impedance. There's no way to match impedance of 8 wires in all possible pair combinations. Ignoring this will greatly reduce signal integrity making connection unreliable.
The image (taken somewhere form the internets) shows proper (not perfect however) PCB trace design between a jack and a transformer. Traces are grouped by pairs corresponding to each twisted pair in a cable.
@iliazeus Wait, do you mean for each wire or each cable?
If cable, then yeah, you need to save wiring state for each cable, but then again, that's what, around 8 kilobytes total for a 52-port switch? Man, I've seen ARP tables much bigger than that.
@drq I doubt that random pair of network adapters could work via randomly inserted A-B cable. Most of cables are A-A, some are B-B. If wire order is same, their colors don't matter
@drq again, 2 combinations is not 8! combinations, and I can imagine analog scheme for doing it. Set state once, and work forewer
...but this is not easily scalable to 8! possible combinations
@drq go fucking replace all network adapters/routers/motherboards, but before that, try to convince business it requires to buy new adapters/routers/motherboards which are 10% more expensive than normal
@mo hardware gets replaced, like, all the time, gal. Just naturally. And also, new houses and workplaces need wiring as well, and the easier and quicker it is, the cheaper it will be. Human time is way more expensive than machine time. Your argument sounds like "screw your automatic transmission, we don't want to replace our manual gearboxes". And this argument was made, multiple times.
But look around, automatic transmission is everywhere at this point. So manual cars did get replaced by a more convenient alternative eventually. Same with AutoMDI-X. That's why you don't see cross-wire cables anymore, because it was stupid.
@mo@drq But I still fail to get, why on the fly and why transmission speed matters? What we are talking about is just one time procedure upon physical layer connection.
Add comment