[Soekris] Strange problem with net4801

Poul-Henning Kamp phk at phk.freebsd.dk
Thu Jan 20 21:03:38 UTC 2005

In message <20050120174125.GE22898 at ref.nmedia.net>, Chris Cappuccio writes:
>Horst Laschinsky [horst.laschinsky at physik.uni-erlangen.de] wrote:
>> Well OK. Now I understand the fact that shielding on both ends
>> sometimes _causes_ line noise instead of avoiding it. OK, thats
>> pretty clear now. But... :-)
>Talk to anyone in telco and they'll tell you that you can easily have the
>same problems (if you ground both ends) with shielded E1 cable bundles, DSL
>bundles, etc, the difference in ground potential between devics is a big deal

I saw a 20kA groundloop once, its a sight to behold (from a safe distance).

>> 1.) I still do not understand, how this noise can be triggered by
>>     the application layer protocol (as I said, in my case: scp).
>scp was probably the only protocol you were using to transfer large amounts of
>data in a short period of time and so that's where the problem appeared,
>when you made heavy utilization of the link

It's actually far more likely that it is related to packet length and
bit density.

try using ping with various packet lengts and contents.

If your cable runs through or near high current power cables, you can
have an effect where a change in current in those cables gives a DC offset
in the UT cable (because the twist of the the cables does not cancel
out perfectly)  The DC offset may push the magnetics or input stage of
the receiving end out where not all bit densities can be handled.

In that case use STP instead, the shield takes most of the rap, but
make sure to not DC ground it in both ends:  DC in one and and AC
in the other (if necessary).

Or, to bypass the problem entirely:  fiber.

>> 2.) Why do people sell TP-cables which are shielded on both ends
>>     at all?

Ethernet jacks are not supposed to be zero-Ohm terminated, they should
either be LC or R terminated to prevent ground-loop currents.

Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.

More information about the Soekris-tech mailing list