Bizarre ethernet problems (Interlan & 4.2 BSD)

[Rob Warnock]

+---------------
| I've been experiencing strange ethernet problems. Maybe someone out
| there can help me out. First let me say that we have 3 750's all
| running 4.2 BSD with Interlan's controller.
| ...ihnp4!tellab1!steve
| Steve Harpster
| Tellabs, Inc.
+---------------

Incidently, whose transceivers are you using? Interlan? DEC? TCL? (Other?)
What version of the spec (1.0 vs 2.0 vs 802.3)? They all interoperate (usually)
on the cable, but the problems/symptoms on the controller side are different.

+---------------
| 1.	When a tap is unplugged from the controller, then plugged
| 	back in, the system reboots. No error messages, panics,
| 	nothing --- it just starts rebooting. I can live with
| 	this but it's not too friendly.
+---------------

I have seen a note on this before... the transceiver may (per spec) draw
up to 1/2 Amp of +15 volts (actually, 12-15 +/- 5%), which the controller
provides from its host computer's bus, in this case your Unibus backplane.
Now a half-Amp is a good bit, especially if the transceiver has heavy
filters (capacitors) which cause a heavy startup surge (NOT spec'd, although
conservative design would suggest not drawing more than the allowed 1/2 Amp).
Plugging in the transceiver can therefore cause a temporary overload on the
+15v supply (of some microseconds or even milliseconds), which can cause
the power-fail detectors in the VAX (which are QUITE good!) to trip, forcing
and immediate and complete power-up reset sequence! (So that's why you don't
see any console messages...)

If you have people handy with electronics around, they may be able to
build you a little surge filter (with a power supply "choke" inductor
or equivalent, plus a swamping resistor) that you can put in series
with the +15 line to the transceiver (pin 13), as close as possible to
the controller board. If the D.C. resistance of the choke is low enough
(a few ohms) and the inductance high enough (a Henry or two), it should
cure the problem.

WARNING: I have not tested this and cannot guarantee the results. (You may
have problems with the inductance causing an instability in the D.C.-to-D.C.
converter in the transceiver. That's why I mentioned the swamping resistor.)

+---------------
| 2.	Because of an incredibly poor design, the cable from the
| 	controller to the tap quite often comes loose. This is
| 	noticed because the machine appears down and the lights
| 	on the controller stay dark. The problem is, none of the
| 	other machines can talk either...
+---------------

If this is true, the transceiver is not meeting spec. I'm guessing you
may be losing one side of the transmit pair, causing the transceiver to
think you are sending continuously, but the transceiver's "jabber control"
circuit should stop this after a few (20-150) milliseconds. Make sure your
transceivers are all equipped with "jabber control". If they aren't, complain
(and replace them!). (In the version 2.0 Ethernet Spec., this is called a
"Transmit Watchdog Timer", Section 7.4.6)

+---------------
| 3.	This is the most disturbing. Every machine appears down to
| 	the others. The taps are secure. Each controller on each
| 	machine periodically flashes the carrier sense light. It
| 	looks like they each detect activity on the network but
| 	they can't figure out who's doing it.
| 
| For the most part, the ethernet works great. Whenever we do any
| activity under the floor though and bump the cable everything goes to hell.
+---------------

Check the mechanical integrity of all joints, the physical taps, and the
termination resistors at the ends. A short circuit or an open circuit
will cause each transmitter to think that it's "colliding" with someone
else (due to reflections from the "mismatch"). If you have controllers equipped
with the TDR (Time Domain Reflectometer) feature (counts "ticks" from transmit
til collision), you can use that to narrow the source of the problem to
within a few tens of meters. (A real TDR can be used if you get desperate.)

Hope that helps.

Rob Warnock

UUCP:	{ihnp4,ucbvax!amd}!fortune!redwood!rpw3
DDD:	(415)572-2607	(*new*)
Envoy:	rob.warnock/kingfisher
USPS:	510 Trinidad Ln, Foster City, CA  94404	(*new*)