On Wed, Nov 13, 2002 at 11:40:11PM +0100, Manuel Bouyer wrote:
> On Tue, Nov 12, 2002 at 03:51:05PM -0800, Andrew Gillham wrote:
> > 
> > In debugging my IPv6 issue I discovered that tcpdump on the epic0 in my
> > main server causes the interface to reset, or at least hang for some time.
> > Open connections to the box usually timeout before this completes, but not
> > always.
> > 
> > It is on a NetBSD/i386 -current (October 28) system.  Has anyone else seen
> > something like this on an epic?
> 
> I usually see a hang of a few seconds in network traffic, when
> turning on/off promisc mode. You say connections timeouts, how long does
> it hang ? 

Ok, so the hang is about 35 seconds.  I looked on my switch and the port
is actually bouncing, presumably due to a reset when going into promiscuous
mode and back out.

Since this port is configured as a trunk on my switch, spanning tree was
blocking the port for around 30 seconds after the bounce.  I was able to
disable spanning tree even though it is a trunk port.

Is it absolutely necessary for the link to bounce in this case?
Can I force the card into promiscuous mode all the time?  It is on a switch
port, so traffic shouldn't be a serious problem.  I don't usually have any
unicast flooding on my switch.

-Andrew