What happens if you somewhat exceed the 100m limit? The only sources I can find boil down to just don't do that but I can't find a chart showing what happens to performance/frame loss.
> What happens if you somewhat exceed the 100m limit? The only sources I can find boil down to just don't do that but I can't find a chart showing what happens to performance/frame loss.
It depends on your system. If you're running old school shared medium Ethernet (with hubs, not switches, for cat5), or even just half-duplex mode, if your run length is too long, the big issue is that you've lost guarantees about collision detection; in that case, receivers may see collisions that senders don't, and you're left with transport layer resends (if any) instead of Ethernet layer resends. This is because of transmission speed, minimum packet lengths, and the length of the cable mean that one sender could finish transmission before seeing the second sender's preamble.
If you're running full-duplex dedicated medium, collision detection is a non-issue, there are no collisions. GigE transmits and receives on all four pairs simultaneously and uses echo cancellation to remove the echo'd send that comes back from the far end, the extra distance will make the echo come back later than expected, and that might be outside the ability of the network cards' signal processing to adjust.
Assuming that's not a problem, you would have more resistance, and probably more? capacitance, which may make signal recovery more challenging. 100BaseTx would probably be more likely to work than GigE, in my opinion, but I'd certainly try both. Ethernet autonegotiation runs on a single pair at 1Mbps, so it's quite possible that the cards will negotiate to 1G, and then not be able to actually communicate at that speed because of the wiring parameters.
One important thing to consider is that Ethernet wiring specifications are such that minimum standard wiring will work at a very low error rate at maximum length when in a bundle with many other ethernet cables/other low voltage signaling cables. If you don't have a lot of bundled cables, you have a bit more wiggle room. Of course, if you run the wires parallel and close to AC wires, you have a lot less wiggle room.
Oh, and one more thing. If there's a midpoint somewhat accessible, there are two-port PoE powered switches you can drop in the middle of the span; some of them even will chain PoE; I think i've seen something that would let you put in three of the chainable switches between the power source and your endpoint (you could probably chain a couple more if the endpoint is also a powered switch).
It would certainly make network engineers cry, but you have to work with the site and wiring you have, and only sometimes is it easy or convenient to run new cabling.
The 100m eithernet distance restriction used to be to ensure that if two machines tried to start talking on the same line, they would hear the noise from the other before they finished sending their packet, and could therefore know there was a transmission error.
That is, at least, what was described to me in Computer Networks I back in 2006. I'm sure things have changed as we've moved to always using full-duplex wiring and fully-switched networks.
It is really about the modulation scheme limits (distance v noise). Most ethernet networks have been full-duplex on an individual link level since the early 2000s.
It depends a lot on cable quality, what you have on either side and the connectors in between.
At work we had to repurpose a pair 170ish meter cat 6 cables for some cameras and it has worked nicely for three years (with PoE even). We tried again with a 100m cable rated for exterior (plus female + patch cords) and it wouldn't even link at 100mbps
