Re: Doubling 100MBit ethernet by splitting the cable <-- bad idea
>On Tue, 27 May 2003 05:40:06 +0200, J F <j---@---e.com> wrote:
>>>Splitting the cable
>I've seen a network set up with most of the cables were doubled up to
>carry 2 10/100 connections over one wire. I don't think any of the
>computers in the office were connecting at 100mb and the network had
>major latency problems even at 10Mb.
I've run two full-duplex 100BASE-T links through 25 meters of
four-pair Category 5 cable and it worked fine.
I suspect in the office where it didn't work so well, there was
a wiring error, such as not keeping the pairs together or
running a lot of unconnected pairs around.
10[0]BASE-T NICs talk on pins 1 and 2, and listen on pins 3 and 6.
A good cat-5 cable for 100BASE-T either leaves pins 4,5,7,8 completely
unconnected (two-pair crossover cables are made that way), or connects
pairs 4-5 and 7-8 (the standard four-pair straight-through cable).
A good NIC terminates the unused pairs to keep them from resonating.
If you make your own cable it should not have any unconnected pairs
in it, and shorting them is as bad as leaving them open.
I've seen cables for sale at computer stores with wire pairs
on pins 1-2, 3-4, 5-6, and 7-8. I don't know what those cables
are for, but it isn't Unshielded Twisted Pair (UTP) Ethernet.
They may work in some cases, but they'll be impaired becase the
receive pair isn't a Category 5 hundred-ohm transmission line.
Its impedance is too high and its common-mode rejection will suck.
Maybe the PHYs will autonegotiate 10BASE-T to try to get by
the impairment, but they shouldn't even do that.
These cables will buzz out fine with a DC ohmmeter. If you buy
a "cable tester" that says these cables are okay, ask for your
money back. A cat-5 tester should tell you whether each pair is
within 3% of 100 Ohms, and a cable where pins 3 and 6 are in
two different pairs should fail that test.
Cameron
Reply to: