Why would a too short Ethernet cable cause network problems?

[NiHaoMike]

I was having some strange network problems with my PC, from taking excessively long for the link to come up to slow network performance. There is an 8 inch or so patch cable going from the onboard NIC of the embedded Atom platform to the internal switch. When I replaced that patch cable with a 3 ft patch cable, everything worked perfectly. Thinking that cable is defective, I connected it to a network analyzer and no matter how I flexed the cable, it always tested fine. If I use that cable to connect some other Ethernet devices together, it works fine. Clearly, that onboard NIC (RTL8111 based) doesn't like a really short cable.

Why would a cable that is too short cause problems?

[prenato]

If a transmission link is affected by reflections bEcause of bad impedance matching, a short link may be worse than a long one ( reflections are more attenuated)

[Psi]

You could try force the speed to 100mbit instead of auto.
That can help sometimes.

[mariush]

Did you also move the Atom system further away from the switch when you changed the cable?

Maybe it was some interference from the switch causing problems with the Atom mainboard...

[HackedFridgeMagnet]

Try a different short cable

[timcki]

I was using 12 inch cables  and was having problems with connectivity and packet loss.
Switched to 24 inch and problems went away.

they were connected to both a switch and a PC

I do not know why, nor could I find a credible answer why..

[NiHaoMike]

The boards remained in exactly the same place.

I found another short cable that is about the same length and I get the same problem. So I'll just use the 3ft cable and find a way to neatly route it. (Actually, with all the work that was going on, it's a little messy and all the cables look like they need to be carefully rerouted.)

I suspect the equalization circuits are not expecting such a short cable (does Gigabit have a minimum cable length specification?) and/or the receive amplifiers are being overloaded. I found the datasheet for the RTL1111 and there's a 3.3V supply and a 1V supply. Maybe the receive amps are powered from the 1V rail and cannot accept strong signals.

Such an issue would be quite problematic in embedded Ethernet, where the signals run on one board or several boards connected by connectors (like NICs on plug in cards with a switch in the backplane) without going through any cable.

[prenato]

The problem coul be in the transmitter as well. A transmitter with bad impedance matching (ie poor transmit return loss) will not attenuate reflections from the receiver end much. These in turn go back to the receiver and interfere with the signal quality. In these cases, a longer cable actually helps because the reflections are much more attenuated by the time the reach the receiver. Most high-speed transmission standards have transmit return loss minimum specs for this very reason but a cheap transceiver/transformer or board design can result in reflection issues.

[gxti]

Before gigabit there was a minimum length requirement for some reason or another, my guess was overloading the receivers but it could also be impedance related.. It would usually work but there was no guarantee. But gigabit got rid of the minimum length, so in theory that shouldn't be a problem. Still, reality and specifications don't always agree and some combinations of transceivers might not work with short cables.

[grumpydoc]

Before gigabit there was a minimum length requirement for some reason or another,

For collision detection IIRC