[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: More 100Mbit



On Sun, 14 Jun 1998, Doron Shikmoni wrote:

> >Pardon me for barging in, but did you know that the usual flat cable used
> >in 10 MBPS ether with RJ 45 connectors is NOT approved for 100 MBPS ?
> 
> The cable for 10BaseT Ethernet is *not* flat. It is made of Twisted
> Pairs: each *pair* of leads is twisted within itself. It is the most
> important attribute of the cable.

Yes, but the 2 twisted pairs are usually molded in flat plastic envelopes
when used with RJ 45. Thus people confuse them with the modular wires
sometimes used for short connections on 10 MBPS (and these are not twisted
- max length 1.5 meters).

> Indeed, the requirements for 100Mb/s are higher than those for 10Mb/s.
> The commonly used standard is EIA/TIA-568A, and it categorizes the cables
> into 5 categories; only two of them (3 and 5) are actually relevant
> when we talk about Ethernet and Fast Ethernet. 10Mb/s can safely run
> on cat. 3 cables; 100Mb/s requires cat. 5.

You have the standards handy ;) The cat.5 wiring is mentioned everywhere,
but reading through that standard will upset anyone's stomach. Could you
perhaps quote the essentials ?

> >If you happen to hit a (sub) multiple of lambda with your cable length,
> ... 
> There are many more parameters than the lambda multiples. It's far
> more complicated - beyond the scope of this list (we're offtopic
> anyway).

The point I was trying to make was, that unmatched cable will work over
short distances, IFF not a whole multiple of lambda/4 at the baseband
modulation speed. For wiring under 1 meter, I have even used twisted phone
pairs. (obviously for testing only), and I have used 'link' cables in 900
MHz circuits made in the same way. 

> >more so). If I remember well, they want 120 ohm cables (but I may be
> >wrong).
> 
> 10BaseT and 100BaseTX are specified for 100 ohm cables. Not 120.

Point taken.

> Ethernet was specified for a coax cable), however the signalling is
> totally different and you need a different transceiver (either built
> in the NIC as in "combo" NICs, or an external one).

I always try to buy combos. The extra price returns tenfold by removing
pains with wiring and hubs.

The signalling is not that different, and it is a sin to use that many
wires for a signal that fits confortably in the 1 GHz + bandwidth of a
simple coax cable (that costs $2/meter or less).

Anyway CSMA/CD necessarily goes down the bog when load goes over 50 %,
unless used point-to-point, no matter how many wires you use.

BTW: Is there some progress with Token Ring over Ethernet for Linux ?
There was some talk once, to use a Token Ring type Tx sheduling on a
normal Ethernet, to make it behave better under high load. Anyone ?

> >120 ohm 1pair + shield coax
> >was used by IBM token ring ? Maybe you can find an old cable and get the
> >wire from there.
> 
> The IBM Token Ring does not use any coax. Its cable is two-pair, twisted,

Well, something did use such a cable, and I had one once. It had a special
connector like BNC but with 2 pins inside it. And it was 120 ohms (I
measured). I thought it was some form of old ethernet or token ring.

Peter