tulip driver & Asante Fast 10/100 card

[Bob Farmer]

> 
> > If you are using half-duplex on either end (ie these combinations:
> > half-duplex on both ends, half-duplex on machine and full-duplex on
> > switch, or full-duplex on machine and half-duplex on switch), it never
> > locks up.  If, however, you are using full-duplex on both ends, it always
> > locks up once you try to receive a large file with FTP.  If, again while
> > both ends are at full-duplex, I try to _send_ a large file with FTP, I get
> > this (also easily reproduceable):
> 
> It might be worth going over your cabling with a fine-tooth comb, and
> checking for any cable segments which are improperly wired.
> 
> I've seen a _lot_ of "Cat-5" cables wired up wrong.  Many cables (both
> manufactured in-house, and commercially purchased) simply connect
> the four Cat-5 pairs sequentially across the pins of the RJ-45
> connector.  This is a violation of proper wiring practice, because
> the 10- and 100-Base-T signals are _not_ paired up in this fashion.
> 
> I've even seen some facilities wired up wrong - the eight conductors
> in the Cat-5 cable were punched down into the 110 blocks in this
> straight-across fashion.
> 
> If you try to run Ethernet over this sort of cable setup, bad things
> can happen.  You'll be sending a signal which _should_ be travelling
> along one specific tightly-twisted pair, over two wires which are
> actually in two different pairs.  This will mess up the characteristic
> impedance of the cable (leading to possible signal reflections),
> cause the cable to radiate more EMI than it would otherwise, and
> leave the signal more prone to external interference, crosstalk,
> and so forth.
> 
> I'd expect that full-duplex transmission over a miswired cable would be
> particular vulnerable to problems, because both pairs are active
> simultaneously.  If you've got any sort of crosstalk problem within
> the cable, full-duplex transmission could really exaggerate it.
> 
> There are even worse sorts of miswiring possible (and I'm sure that
> most of them have occurred in one place or another).  Things would
> get really messy if, for example, someone managed to "cross pairs",
> so that parts of the TX and RX signals were accidentally being carried
> by wires in the same twisted pair.  You'd have unbelievable crosstalk
> problems in that case!

Well, I'm fairly confident the problem lies in that particular card's
interaction with the tulip driver, and not cabling or other external
issues, because when I switch out the Asante PNIC card with an older
Asante card that uses a real DEC 21141A, things work perfectly and at full
speed at any duplex.

-- 
Bob Farmer                                     ucs_brf@shsu.edu
Computer Services, Sam Houston State University; Huntsville, TX