Hacker News Reader

I hate downvotes. I usually upvote when I think someone has been unfairly downvoted. I very rarely downvote, and yet I was one of your downvoters.

It's clear that you're "ignorant" of the black magic involved in high speed communication. I felt that your post amounted to "I don't understand this stuff ... therefore it is probably a scam".

As others have pointed out to you, at high enough speeds things become very analog instead of digital. You can literally fit entire Ethernet packets into a twisted pair cable. Packets can be short enough that they exist completely "in the wire". That's a lot different than what a "premium audiophile cable" does.

If you ever want to learn just how much "black magic" there is, read Howard Johnson's books.[1] The information is somewhat dated, nowadays things are even weirder.

[1] https://en.wikipedia.org/wiki/Howard_Johnson_%28electrical_e...

It's clear that you're "ignorant" of the black magic involved in high speed communication. I felt that your post amounted to "I don't understand this stuff ... therefore it is probably a scam".

"ignorant"? That's really jumping to conclusions...

I've worked with DDR, DDR2, PCI, PCIe, and USB (2.0 only, but that's still 480MHz), in mass-produced designs. Also some proprietary busses operating in the 600-800MHz range. It doesn't have to be perfect. That is what is so great about digital signaling.

My post is more of a "I know from experience how much you can get away with, and fancy test equipment that can tell the difference does not always reflect how something performs in practice."

On the other hand, I won't comment on high-frequency true analogue stuff like microwave/RF.

I see the point that you and 'parennoob' are making. The word "ignorant" was a poor choice on my part.

It is a bit funny that your verdict is that someone is "ignorant" of the black magic in high speed communication, and you're linking to someone who literally has called two of his books "Black Magic".

Now I don't happen to agree with 'userbinator's skepticism -- indeed I think that a lot of engineers who work with digital stuff tend to forget that it is analog under the hood, with all the associated crosstalk and BER problems. But creating this atmosphere of "Ooooooh, it is something you probably don't understand" isn't very productive, if you know what I mean. I wish Johnson had titled his books, "High-Speed Digital Design: Quite Simple If You Do the Math and Physics".

I see the point that you and 'uberbinator' were making; "ignorant" was a poor choice of words on my part.

I intentionally used the words "black magic" because I was going to link to Johnson's books. While Johnson's publisher (not Johnson) probably named the books, I like the titles. The words you suggest may be more appropriate but the actual titles are definitely "catchier".

You'd be surprised at just how analog high speed digital is. The data represented by the signal is digital, but the signal over the wire is an analog waveform more complicated than simple high and low voltages. Poor cables can absolutely cause high BER and lower performance. If you've ever watched digital television over the air and seen the picture drop out, it's a similar thing.

A lot of the audiophile nonsense is only nonsense in its application to the ridiculously low frequency world of audio. But when we're dealing with transmission lines, yes, cables absolutely do matter.

The data represented by the signal is digital, but the signal over the wire is an analog waveform more complicated than simple high and low voltages.

I agree, but as long as the receiver interprets the values correctly (i.e. the 1 is above its threshold, and the 0 is below its threshold) it doesn't matter. Among other things I've watched what signals like USB 2.0 HS really look like on an oscilloscope, so I'm well aware that they don't look anything like the nicely-drawn diagrams in textbooks.

Poor cables can absolutely cause high BER and lower performance.

But was that the case here? They identified a cable that had sub-par signal characteristics, and without looking at the actual effects it had on network performance, made the premature conclusion that it was. We don't know whether or not it was really the cause of the customer's problem (imagine them getting back this nicely terminated cable and seeing that it has the same performance... I would not be surprised.)

> [Blue Jeans Cable] identified a cable that had sub-par signal characteristics, and without looking at the actual effects it had on network performance made the premature conclusion that it was.

From the fine article:

"[Our customer is] dealing with network performance issues at a small company, and has come to suspect that bad patch cords, made by an electrician, account for some of the problems he's seeing."

Looks like the customer who sent them the cable had already done the troubleshooting required to determine that the cable was probably bad.

Also, you should probably read: http://www.bluejeanscable.com/articles/channel-certified-eth... Each part in an Ethernet network has well-specified tolerances for a reason. If one part of that network causes far too much signal degradation then all sections of the network that flow through that part can fail to function.

> without looking at the actual effects it had on network performance, made the premature conclusion

They didn't need to look at the "actual effects".

The very smart engineers who created the high speed Ethernet standards have already studied the "actual effects". They spent years designing, studying, analyzing, and characterizing.

The designers distilled their efforts into some simpler rules such as "near end crosstalk (NEXT) shall not exceed ...".

Those simpler rules allowed a company like Fluke to build a test instrument that checks for things like crosstalk and delay skew and report PASS and FAIL and even report how far off the cable was from meeting the requirements.

That's all there is to it. It's ludicrous to ask for an ab initio analysis of every network problem that could be possibly encountered.

It may eventually be necessary to study "actual effects" in a particular network, but only after dispensing with low hanging fruit such as poorly terminated cables.

There is also the long term labor costs of breaking what amounts to a standard API.

Perhaps this mfgr card will work on a almost-but-not-quite cat5 cable run, but that new one won't, or that other one only works half the time, and slowly, or that one won't work when that other fluorescent light is on. The labor cost of this kind of troubleshooting can be pretty spectacular compared to a cheap and simple "must conform to cat-5 minimum standards".

Its rather like AC line voltage. True, some stuff, sometimes, will work at 100 volts or 150 volts, but I'd call an electrician and get it fixed before you spend a lifetime troubleshooting individual things that won't work reliably and consistently at 100 or 150 volts.

It is similar to DSL connection issues cause by spotty telephone wiring. Cross talk and outside interference raise noises levels on the line and the devices on either end can't tell the signal from the background noise. This is different from audiophile cabling, which is usually over short distances where noise is not an issue.

The difference between analogue and digital in signalling is that with digital, if the signal integrity through a cable reaches a certain threshold, then you achieve perfect transmission of whatever protocol you are sending. Where this threshold is depends on how much bandwidth the protocol uses.

This does not mean that all digital cables are fine then - with very long, low quality and/or improperly terminated network cables, you tend to see higher packet loss, sometimes you see dropouts, and sometimes the equipment negotiates to a lower speed. This kind of thing has a lot to do with what's happening in the analogue domain, because interference and signal integrity problems interferes with decoding the analogue signal back into the digital data, so you have packets that fail the checksum validation because they have been corrupted mid-flight.

The problem is that what we send down cables aren't nice sharp square-edged waveforms. If you decompose a square-edge digital waveform into its Fourier series, you'll find that it takes roughly 10x the fundamental frequency of the digital rate to decently represent the signal -- so a 1 Gbit/s digital signal would take 10 GHz of analog bandwidth. Naturally this is a big waste from a Shannon-Nyquist perspective, so in practice what gets sent aren't nice sharp transitions, but rather curvy ones without so much high-frequency content.

It's still true that there are voltage thresholds at the receiver, but for high-speed digital comms there are now _time_ thresholds. In a transmission line, discontinuities cause partial signal reflections that can show up at the receiver to mess up your day. A more intuitive way to show cable quality is the eye diagram, which tells you if the threshold and timing requirements are being violated.

http://en.wikipedia.org/wiki/Eye_pattern

Edit: But this is getting away from the point of the article, which is that bad terminations cause crosstalk. There's nothing mystical about cables failing ISO specs for crosstalk. I agree with you though that some before/after packet loss numbers would've strengthened their case.