|
I used to work work for a company involved in 10GE some years ago. I believe maintaining would be no link drops, but would include fast-retrains, where the cable coefficients are quickly synced without losing link, resulting in some number of dropped packets around the event. Low signal-to-noise will result in a higher bit error rate and sporadically dropped packets. Adding aggressors (noisy fluorescent tubes, power lines, RF transmitters, etc) would eat away at the margin. But, if you're doing short runs (like 30m), then you've got absolute loads of headroom, at least with the fairly standard equipment we were testing. And also, the bit error rate is for full speed. If you don't have constant 10G traffic, then your error rate will be proportionally lower since there will be a bunch of "idle" time. When testing these, there's a bit error rate test performed at different lengths of cable. IEEE used to recommend something like 10^-13 (not sure if this has changed), but we would test, with some margin, at the full 100m using cables wrapped on spools (no external aggressors). We would also test with a "5 around 1" configuration, with 5 other fully active cables tied around the cable under test. If I recall correctly, this test required the full 10^-13. All of these cables were periodically verified to meet all specs. But regarding cable quality, we would buy thousands of meters of cable without issues, from good vendors. You may have trouble finding compliant cable from China. If there was a problem, it was usually from someone bending and putting a permanent kink in the cable (causing a discontinuity in the impedance) or closing a door on it. |