Date: Sat, 17 Jul 1999 01:54:06 -0700 From: Tim Baird <tim@storm.digital-rain.com> To: freebsd-hackers@freebsd.org Subject: Re: poor ethernet performance? Message-ID: <3.0.2.32.19990717015406.0095f670@storm.digital-rain.com> In-Reply-To: <Pine.BSF.4.05.9907170021220.331-100000@venus.GAIANET.NET> References: <3.0.2.32.19990716231622.007e2100@storm.digital-rain.com>
next in thread | previous in thread | raw e-mail | index | archive | help
>> >> Now for a brief tutorial on transmission line / gigabit ethernet...... >> >> >> Ensuring that the entire transmission path maintains a consistant >> characteristic impedance is the most difficult task in cable manufacture. >> It has typically the most influence on the quality of the signal >> transmission. Crosstalk is always present in a multi pair cable system, it >> is a matter of degree i.e. how much cross talk (undesirable signal) is >> present as a percentage of desired signal. The level of crosstalk is >> proportional to dv/dt .... the rate of change in signal voltage with >> respect to time. This is different than "frequency". A 1 Hz signal that >> approaches a square wave (or rectangle) can have HUGE crosstalk at the >> edges because of the large dv/dt at those points. Care must be taken to >> optimize the dv/dt with respect to the desired baud rate. (baud == state >> changes per second) >> >> As a cable length increases, losses increase, (these can be compensated for >> by increasing drive level ... and thus dv/dt) but a potentially worse bogey >> man plays a role....propagation delay. Most of the physical problems >> previously indicated can be improved upon, but no one has found a way >> around the limits of the speed of light (most transmission line allows >> propagation at about 70 to 80% of c). The time is not far off where this >> will be by far the most significant limit on information exchange for >> everyday communication. >> >> STP is better to use over shorter lengths where high level of EMI >> compromise the common mode rejection of the reciving system. The downside >> is that STP typically has rotton characteristic impedance >> consistency....not because of the plastic jacketing etc, but because of the >> varying distance (radius) between the cable pairs and the sheild over the >> length of the cable. Going to coax is usually the choice here for standard >> ethernet. >> >> Gigabit ethernet obviously creates a new level of awarenes about all of >> these factors. From a great article at >> http://www.gigabit-ethernet.org/technology/whitepapers/gige_11.97/how.html.. >> ..... >> >> * Use existing 4-pair Category 5 cable. To ensure proper operation at full >> link lengths, the cable must conform to the requirements of >> ANSI/TIA/EIA-568-A (1995). >> >> * Use all four pairs in the cable to keep symbol rate at or below 125 Mbaud. >> >> * Use PAM-5 coding to increase the amount of information sent with each >> symbol. (similar concept to analog modem methods) >> >> * Use 4D 8-state Trellis Forward Error Correction coding to offset the >> impact of noise and crosstalk. (ie. reciever has the ability to correct the >> error without requesting retransmission) >> >> * Use pulse shaping techniques to condition the transmitted spectrum. (i.e. >> limit dv/dt etc) >> >> * Use state-of-the-art DSP signal equalization techniques to manage the >> problems of noise, echo and crosstalk interferences, and to ensure a bit >> error rate of 1 x 10 exp(-10). > > Thanks for the article and for the brief. I just have a little >comment on shielded versus unshielded for both analog and digital audio >cables, not sure if this applies to data cable but digital audio is data: > >Cables are of the "nude" (unshielded) style, which has generally been >perceived to sound "faster" and less "colored" than conventional fully >shielded cables. Ahem....I am not sure what this means......It sounds a little bit like those articles that you read in that audiophile magazine "The Absolute Sound" where people claim that they can hear the difference in the acoustics of their listening room depending on where they have situated their teacup......but I digress.....see below.... As it turns out, there is good reason to think so since >properly designed, un-shielded cables are much less reactive to the signal >than their fully shielded counterparts. At audio frequencies and with >reasonably short lengths of cable, a shield typically does more harm than >good and is otherwise necessary only for Radio Frequency transmission >and/or into extremely high gain inputs such as microphone and phono >pre-amps. Instead, properly braided or twisted conductors effectively >reduce susceptibility to induced noise, especially inductively coupled >interference (EMI) while angular crossing weakens the field effects of >opposite polarity conductors on each other. The mechanism for the >self-shielding/field controlling design is to divide the signal into >several separate runs in a continually changing orientation such that only >a small fraction of either polarity is ever in the ideal orientation to >the wave front. This has most relevance to electromagnetic fields either >internal or external, which especially require an optimal angular >component to induce the greatest opposing current flow. This is not a very accurate description of what is happening in a twisted pair situation....... There are two factors influencing the design of twisted pair transmission line. First there is the need to control the consistency of the characteristic impedance of the line. This happens by controlling the distance (radius) between the two conductors as well as the dielectric material that is present in their vicinity. For coax, this is a special compound between the braided outer conductor (shield) and the inner conductor. For twisted pair, this is usually air (the dielectric properties of the insulation are made to be virtually the same as air). The dielectric only affects the capacitive element by controlling permittivity, the inductive element is essentially controlled by the permeability of free space. The characteristic impedance has nothing at all to do with the DC resistance of the conductors used...it (CI) is a combination of the two reactive effects inherantly present along the line...namely capacitance (storage of electric charge) measured in Farads/meter and inductance (storage of magnetic energy) measured in Henries/meter. These two effects or "reactances" will always be equal and opposite at some frequency, the idea is to have an effective cancellation happening over as broad a range as possible so that they do not limit the bandwidth of the transmission line. The values of the two reactances are positive and negative imaginary numbers. When you take the square root of the ratio of the two values, you end up with a real number in the unit of "ohms".....ie the characteristic impedance......... BTW This is EXACTLY analogous to "index of refraction" with respect to light....... As the spectral components of your signal increase in frequency, there are some interesting effects which occur...one is "skin effect". This effect is the tendency for currents on the line to travel to the outside of the conductor as their self induced electromagnetic force creates a higher flux density at the center of the conductor...this changes the effective cross-sectional area of the conductor and thus the resistance. The result is a more "lossy" line....Other losses are found in the dielectric material, and of course the natural resistance of the copper.... Second, to minimize the effect of electromagnetic interference (EMI) which is both "crosstalk" and other stray emag signals incident on the transmission line, the twisted pair design is used on the assumption that common mode signals are effectively rejected by the reciever...ie any currents induced along the line that are EQUAL will be cancelled out by the differential reciever. If you twist the cable pair as it travels through space, you will have any incident signal affect BOTH conductors equally (hopefully)....thus the EMI induced currents will be effectively cancelled out by the commond mode rejection characteristics of the differential receiver....This is why you twist your antenna wire that comes from your TV antenna on your roof (if anyone still uses these)...otherwise your TV will have a ghosted image...one signal from the antenna, and another signal (shifted in time) coming in along the antenna wire. As for these issues in the audio realm, I can assure you that high frequency effects such as skin effect, characteristic impedance considerations, and the like have NO real influence on your sound quality...anyone who says so is shoveling cow cookies, or has a high bit error rate in the cerebral cortex For input stages like microphones etc, there is the need to block out EMI so that you don't record your guitar along with the 60 Hz hum from your electrical system, the brush noise from your neighbour's drill etc.....This is where using a combo of sheilded and twisted cable is preferred....... As for data transmission, you want to maximize the data rate, maximize the distance, minimize the error rate, simplify the installation procedure, minimize the cost etc....Therefore, you want to have all of these design considerations outlined above built in to the whole system. Also, you don't want to have to rip out the network cable every time you upgrade your data pumps...You want as good a cable as you can afford, that way the data pumps (ethernet cards) are able to operate at their full capacity without wasting time doing retrans....Most of the time though, a reasonably conscientous cable installation will not be found to be the weak link in your system. Usually it will work great or really badly, and the cause - if it is bad cable - is usually found easily by isolating different segments and testing.... I hope everyone is benefitting by these simple facts....I hate to see needless misunderstanding BTW, if this is too far off topic for this list, please feel free to express your concern (anyone) To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3.0.2.32.19990717015406.0095f670>