Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site fortune.UUCP
Path: utzoo!watmath!clyde!floyd!harpo!ihnp4!fortune!rpw3
From: rpw3@fortune.UUCP
Newsgroups: net.lan
Subject: Re: Re: Re: Ethernet cable length query - (nf)
Message-ID: <2895@fortune.UUCP>
Date: Thu, 29-Mar-84 04:03:54 EST
Article-I.D.: fortune.2895
Posted: Thu Mar 29 04:03:54 1984
Date-Received: Fri, 30-Mar-84 02:35:39 EST
Sender: notes@fortune.UUCP
Organization: Fortune Systems, Redwood City, CA
Lines: 86

#R:watcgl:-232100:fortune:5900009:000:4713
fortune!rpw3    Mar 28 22:01:00 1984

In fact, MOST of the Ethernet chips have some sort of TDR register, and
where they don't, it is easy for the designer to add one externally (as,
for example, outside the Seeq chip).  They aren't really "Time Domain
Reflectometers", but the effect is similar.

The way they work it quite simple, but has a irritating limitation.
The "TDR" register simply counts the number of bit times from the start
of transmission until the collision-detect signal comes back from the
transceiver. Simple. If all your stations report continuous "failure to
send after 16 collisions", you probably have a solid fault in the cable
somewhere, and the "TDR" numbers will tell you how far from the station.
Get "TDR" reading from two stations to resolve the ambiguity of which
direction down the cable has the fault.

Now the catch. Since they are counting transmission bit times, the
resolution is about 100 nanseconds, or about 20 meters (65 feet).
And since the collision-detect signal is normally generated by an
unsynchronized clock in the transceiver (yes!), that doubles the
uncertainty, to 40 meters or so. But since they are measuring round-trip
times, the actual uncertainty is half that. (Got that? Back to 20 meters...)
Since transceivers may be spaced as close as 2.5 meters, that's still
a lot of poking around up in the ceiling!

What you need to do is attempt to transmit from a given station several
times, and take the LOWEST value reported from the "TDR" register. This
will hopefully remove the uncertainty caused by the unsynchronized
transceiver oscillator. This should lower the uncertainty to about 10 meters
(~30 feet). Do this for several stations. Remember to subtract the transceiver
cable length from each station to the cable when converting "TDR" numbers to
distance. There should be no more than four or five taps in the area
thus pointed to. Example:

	Terminator						Terminator
                    <--25m-><-25m-><-10m-><---37.5m--><-5m->
	   R--------X-------X------X--X--X------------X----X--------R
		    |       |      |  |  |            |    |
	+---+	    |       B      C  D  E        10m |    G
	| A +-------+                               +-+-+
	+---+  20m                                  | F |
		                                    +---+

Suppose station A's TDR flickers between 8 and 10, and station F's flickers
between 5 and 7. Then allowing that a count of 8 COULD be as low as 7.001
rounded up, you can assume that the problem is not less than about
(7 * 100ns)/(5 ns/m) ~= 140m or 70 meters one-way from A, and likewise
not less than (4 * 100)/5/2 ~= 40 meters one-way from F. (It could be as much
as 20 meters further from each if you only took a few samples, but it's not
likely.) Looking at the map, you should check out taps C, D, and E.

One further note: Each brand of network controller and each brand of
transceiver will have it's own particular "start-up delay" in both the
transmission path and the collision-detection path, which will result in
a constant (positive) offset in the TDR register values of 4-5 or more bits.
When you are first setting up your network, you should check this out by
forcing a failure (take off a terminator resistor, for example) and noting
the minimum values. (You MAY be able to do this on a transceiver that's
simply not connected to the Ethernet cable, but "clever" transceivers may
be able to detect that condition. If this DOES work, it will automatically
take care of the length of the transceiver cable for you, so don't count
that twice.)

True TDR's (Time Domain Reflectometers) are calibrated oscilloscopes with
fast rise-time pulse generators in them. Depending on price, their accuracy
and resolution can be as good as a few centimeters. Not only that, but you
will be able to SEE all of the good splices and taps between the TDR and the
fault (as little bumps in the trace), so that the fault can be located
relative to the nearest known tap/splice, rather than in absolute distance
from one end. A large site should probably invest in one of them to be kept
in the main machine room at one end of the cable, where the end terminator
can be taken off to do measurements on the (hopefully) rare event of the whole
net going down. MAKE SURE you keep an accurate cable map up-to-date, marked
with all your transceivers and splices in meters from one end.

Having said all of the caveats, the "TDR" register feature IS quite useful
if used properly, and using it properly is no problem once you have made
a few baseline notes. Don't buy a controller without one.

Rob Warnock

UUCP:	{sri-unix,amd70,hpda,harpo,ihnp4,allegra}!fortune!rpw3
DDD:	(415)595-8444
USPS:	Fortune Systems Corp, 101 Twin Dolphin Drive, Redwood City, CA 94065