[Beowulf] Jury rigged ethernet?

Robert G. Brown rgb at phy.duke.edu
Mon Jun 27 08:59:22 PDT 2005

Huntress Gary B NPRI writes:

> This is more than a little off topic but if the answer is encouraging it will strengthen my case to purchase a small-ish cluster significantly.
> Basically I am exploring the idea of placing a cluster in a dedicated
> "non-traditional" physical location where space is at a dramatic premium. 
> This particular space has existing cabling that I cannot change and it is
> my only electrical access.  A single existing cable provides 3 phase
> power, a few DC discrete signal pairs, and a few twisted shielded pairs
> that were used for serial communication.   The cable is approximately 30
> feet long.  I would like to adapt two twisted pairs for ethernet.
> What are the chances that I can coax 10Mbps out of this connection with modern (hopefully tolerant) ethernet cards at either end?  Is this doomed to failure?

I'd generally say excellent -- probably 100Mbps or GigE even, over such
a short run.  I'm guessing that the twisted shielded pairs are cat 3.
IIRC, cat 3 and cat 5 are both 24 gauge wire, so as far as raw
conductivity is concerned you're all right and then some (probably even
if the wiring is REALLY old and is a lighter gauge). cat 5 has a tighter
twist, which basically makes it act more like a wave guide and less like
an antenna.  However, the shielding also prevents crosstalk between
pairs and reduces the possibility of pickup from e.g. the power cables
(presuming that they all run parallel an in a single conduit, a
generally bad idea).

Only 30 feet, though, is a good thing, and with pair shielding I'd
expect thing to likely work find at any speed.

You can find lots of places on the net to find RJ-45 wiring diagrams,




The one thing you'll need to be careful of is wiring pairs as pairs --
only pairs 2 and 3 are used for 100 BT ethernet, but you do want to make
sure that pins 1&2 are a single twisted pair and that pins 3&6 are
another twisted pair (since your color coding is unlikely to match cat 5
-- you may have to test carefully to be sure you have the same cable
pair on both ends if the color coding isn't adequate). If you have the
pairs to spare, you should probably go ahead and wire pairs 1 and 4
through into a standard jack -- you can always split this into a second
channel if you need it at 100 BT, but gigE requires all four pairs.  You
>>might<< get gigE over 30 feet with four separately shielded pairs --
certainly worth the experiment, if it matters to you.

Also, be sure to invest in an RJ 45 cable tester for DIY wiring of punch
blocks, jacks, or homemade RJ 45 crimpled cables, unless you do them all
the time.  When I first started doing my own wiring at this level, my
failure rate was something like 50% -- every other connection had to be
done over.  This rapidly improved as I got the hang of the wire punch
and learned to work meticulously, but you want your connections to be
"perfect" on any link like that this that has other strikes against it.

I am curious, though.  30 feet is easily accessible with a fishtape.
Also, you have twisted pairs already running through the access.  Is it
really impossible to use a TP to pull a cat 5e or cat 6 cable back the
30 feet, or to fish just one more cable into the conduit?  Or is it a
union issue or wiring standard kinda thing?  Having pulled lots of cable
at one time or another over the last twenty years, I'd be more inclined
to sneak a proper cable in than to try to reuse cables that might
generate problems down the road.  A box of cat 5e, a fishtape and a few
tools is a trivial investment compared to the cluster itself... and you
can work in the dark of the night when nobody is around;-)

Once your cat 5 is in, it would be a miracle if anybody ever detected
it, even if they looked.  Who keeps that kind of records?


> Thanks,
> Gary H.
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20050627/f34521e4/attachment.sig>

More information about the Beowulf mailing list