Linux-Hams archive - May 1997: 100 Mbit Ethernet Problem

100 Mbit Ethernet Problem

Mike Bilow ()
Sun, 04 May 97 14:13:00 -0000

Tim Connolly wrote in a message to Mike Bilow:

TC> OK, now I've seen alot of crazy fixes... but could someone
TC> please explain this:

TC> I was working on a friend's 100 mbps Fast Ethernet network.
TC> One of the 40' lines we had ran suddenly quit working. So
TC> we replaced the cable and put new connectors on and it
TC> worked. Except it was pinging about 45% successful and 0% if
TC> you used a larger packet (4K).

TC> By about 1 am, we were desperate. I wound up a couple feet
TC> of cable in my hands making a small 4" coil. Suddenly
TC> packets were flying once again. So we taped the cable, let
TC> go of it and once again, pings were unsuccessful. So we took
TC> a donut looking metal piece and wrapped the cable around it
TC> in a coil fashion. The rate rose to about 75% successful.
TC> Good enough!!!

TC> There was another line which had been run right next to the
TC> original cable. Although it worked perfectly, we tried
TC> disconnecting it and only running one cable at a time. This
TC> had no effect.

TC> Any ideas on why this happened ????
TC> No transmitters were being used at the time. Any ideaS???

Unless you used the wrong cable or have a wiring error, my guess is that you
have a ground loop. This will cause current flow through the balanced line.
The current flow will also be balanced, so the line may still work, but it
messes up noise immunity very severely.

A number of the 100 Mbps Ethernet schemes are far less robust than others. For
example, 100Base-TX will always be touchy no matter what you do.

I assume you are aware that 100 Mbps Ethernet also requires Category 5 cable
rather than the more common Category 3 cable used for 10 Mbps Ethernet. If you
cut open a length of Category 5 cable, you should see very uniform and regular
twists of each pair, about one twist per centimeter. On the other hand, a
piece of Category 3 cable will be twisted at a much lower rate, perhaps one
twist per 10 cm or so.

Finally, check to see that the wires are correctly paired up. If you break a
signal pair across two different twisted pairs, the whole purpose of the twist
is lost and you will see exactly the kind of symptoms you are describing. The
proper pairing on an RJ-45 connector whose pins are numbered in line 1 through
8 is not obvious:

Pair A pins 4, 5
Pair B pins 3, 6
Pair C pins 1, 2
Pair D pins 7, 8

You must also make sure that the lines within individual pairs are not crossed.
That is, make sure pins 3 and 6 on one connector go to pins 3 and 6 on the
other connector in that order. If you cross within a pair, say running pins 3
and 6 on one side to pins 6 and 3 on the other side, then your pairs will still
be perfectly balanced for 10 Mbps Ethernet, but may not work for certain 100
Mbps Ethernet schemes.

-- Mike