2 KNE100TX cards, no hub/switch, options= ?

[Donald Becker]

On Fri, 28 Apr 2000, Eric Jorgensen wrote:

> > > 	Example: Cisco Catalyst switches and 3Com SuperStack II switches, when
> > Sigh.  Those switches are broken.  It is sad that the two largest names in
> But I know what you mean, the right way is the cisco way, eh?
> Even if they change their mind. 

Cisco has a lot of influence.  If someone there screws up a design and
covers it by saying "autonegotiation is bad",  several hundred thousand
CCNEs start chanting "autonegotiation..bad, autonegotiation..bad"

> > > duplex is if the device the AP100 is attached to is already in full
> 	Actually what they said was that it doesn't autonegotiate. And
> they even eventually explained what happens -- 
> 
> "When the AP reboots, we come up briefly at 100meg half duplex, before the
> config file is loaded. When the config file loads, it hits the ethernet
> config record and sets it to full. By this time, the Catalyst has probably
> already switched to 100 half."

The spec says that you have either 1 or 3 (nee 2.4) seconds to respond to
autonegotiation events.  Specifically if you advertise 100baseTx, you have
about one second after autonegotiation completes to start generating link
beat.  If you don't detect 100baseTx link beat in a little over a second,
you fail the negotiation and restart the process.  (This restart is why you
often get 2.4 or 3.0 second cycles of attempts when someone on the link
fails.)

There are two proper behaviors 
  Use an intelligent transceiver that switches speeds for you without
  software intervention.  Before it starts handling traffic the software
  must check the negotiated duplex and flow control.

  Restart autonegotiation by bringing the link down for a full second, and
  then sending the proper Link Code Word.

Story time.  This is pretty close to what the old 3Com 3c595 cards did.  The
initial ones had the EEPROM configured to bring them up at 100baseTx.  The
software was supposed to do the media selection, when the driver was
eventually loaded.

Many 10baseT repeaters of the time interpreted a 100baseTx carrier as a
valid 10baseT signal, with constant collisions.  So when you switched on the
machine your network completely jammed.  In one scenerio, the driver loaded
just before the someone got to the repeater to debug the problem.  In
another scenerio, the driver is never loaded and the entire network is hosed
until the machine is unplugged.

3Com quickly switched the default to powering up in 10baseT mode, but their
initial product run was already out in the field.

> 	So this would leave one to believe that unless you disable
> autonegotiation on the other end of the connection, it will only come up
> 100 hdx. Indeed, when i asked them to verify this, they responded saying
> "Well, obviously!"

"Well, obviously we have broken hardware."

You should use this hardware to be non-autonegotiating.  That means using it
in half duplex only.

> 	And not all RJ45's are created equal. And "Ideal" crimpers are
> less than ideal. I hear ya . . . 

Not all 8P8C plugs are the same.
AMP crimpers have a domed die, with the center elements higher.
Their 8P8C plugs are designed to flex *just so* while being crimped.

Other crimpers have a level die, which is much easier to machine.
Thus they must be used with stiff 8P8C plugs that don't flex as much.

If you use AMP connectors with other crimpers you end up with the center
pins not fully pressed into the wire.

Donald Becker				becker@scyld.com
Scyld Computing Corporation
410 Severn Ave. Suite 210
Annapolis MD 21403


-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to linux-tulip-request@beowulf.org