[vortex] Cable lengths over spec (was: Re: HELP needed: Cross-over connection)

Donald Becker becker@scyld.com
Fri Feb 7 17:22:01 2003


On Fri, 7 Feb 2003, Ruben van der Leij wrote:

> Which brings me on a associated topic. Very long cables.
> 
> One of our locations consists of two building, some distance apart. The
> connection-room (where all outside connections terminate) is in the far
> cornet on the ground-floor, and thanks to Murphy our suite is in the other
> building on the third floor, and front of the building.
> 
> A cable from there to here measures as 118 meters. Which is too long.

Hmmm, I think that you have third-hand information.  With carefully
selected cable, 118 meters can still be within the specs for 100baseTx.

> But 118 meters of fiber and a pair of transceivers is expensive. I assume
> that if I would ask them to put in a cat-5 cable things would work as
> usual.

Ask for the specs of several different "Cat-5e" solid (non-stranded)
cable brands.  Pick the best.  Use short, high-quality patch cables at
each.  It will work, and be in spec.

This is a run you will want to personally inspect for following all of
the wiring rules:
    Pay attention to the minimum bending radius.
    Avoid close parallel runs with any other cables.
    Do not untwist where the pairs are terminated at the jacks.
    Make certain the pairs are not interleaved at the jacks.
    Terminate with a metal punch-down tool that trims the wire end.

> But at which point will things break? How much of a margin is there
> in the specs? (the length-limit has to do with collision-detection, I

It's both a timing and attenuation issue, not directly a collision
detection issue.  The major factor is attenuation, which varies
significantly between cable types.  Both stranded wire and "screened"
cables are much worse than solid UTP, and cables have gotten measurably
better over time. 

Since you are using a switch the cascaded repeater concerns are
moot.  (People used to care about the difference between "class I" and
"class II" repeaters.)

> Any experience on that? And any theory to back that? :)

I used to test with 70-80 meters of abused Cat-2 cable.  It was rare to
find a fast Ethernet transceiver that couldn't handle it.

-- 
Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Scyld Beowulf cluster system
Annapolis MD 21403			410-990-9993