Subject: Re: sysinst network failure (1.5, 1.5.1_BETA, 1.5T)
To: Michael K. Sanders <msanders@confusion.net>
From: Robert Elz <kre@munnari.OZ.AU>
List: port-i386
Date: 05/05/2001 19:28:57
    Date:        Fri, 04 May 2001 23:02:59 -0600
    From:        "Michael K. Sanders" <msanders@confusion.net>
    Message-ID:  <200105050502.XAA09432@shell.aros.net>

  | So, what is sysinst doing differently than ping that could cause this
  | failure?

It reconfigures the network.   This causes an up/down of the net
interface, which then causes many switches to run the full spanning
tree algorithm over the port - just in case it is now connected to
some other switch and is making a loop.

For switch ports that are never going to be connected to anything but
a host, you should figure out how to disable the spanning tree algorithm
(cisco switches use something like "set portfast" or similar).

You really want to fix this, or it won't only be sysinst that doesn't
work correctly, dhclient won't work properly either (it also reconfigures
the interface and then immediately expects to be able to send/receive
traffic).

It is a different question whether most of the interface reconfigs
really should disable the port and re-enable it again (trivial things
like changing addresses, which is all sysinst and dhclient generally do
almost certainly should not).   But reliably detecting the difference
in the driver is not necessarily easy with the current APIs.

kre