[nsp] another counter funny

[David Flynn]

> could be that the switch receives a jam signal from the other side ?

No, this isn't the case.

The collision can only be detected by a transmitting station.  The
collision counter is the number of times a collision has been detected
while this station was transmitting.  802.3 stipulates that if the PHY
asserts a collision while the interface is in full duplex mode it shall
be ignored.  Some PHY's will internally disable the assertion of
the collisions in full duplex mode.

The sign that you are seeing collisions happening on a network is when
you start seeing runts, ie packets beneath the minimum frame size.

A quick note on the JAM signal.  The jam signal isn't a magic identifier
that can be recognised.  Infact it does not matter what data values the
jam signal takes, it is only defined in 802.3 as a sequence of data that
is convenient to the MAC.  What the JAM signal is for is to make damn
sure that the other transmitting encounters a collision.  Without it,
the other transmitting station may never realise that a collision occured.

It is interesting that gert sees an equal number of output underruns as
he does `collisions'.  Since the latter is impossible by definition,
I can only assume that they are connected - it would be interesting to
see if they increment at the same time and under what condition.

..david