Dan Peterson wrote:
We ran into this as well. It seems to be a problem with the E1000 adaptor
(which the 2650 and 1850 [and others] are sporting). Check out
http://www.redhat.com/archives/kickstart-list/2005-March/msg00087.html
From mem doesn't just affect E1000's. Its actually related to the
loader/kernel 'resetting' the NIC interface. The switch, in non-portfast
mode, then goes through its steps of LISTEN, LEARNING, FORWARDING etc
related to standard spanning tree behaviour to protect against bridging
loops in the network. In standard spanning tree this can take upto 30
seconds, although variants do exist to reduce this time, which would
disrupt requests for DHCP information and/or fetching of Kickstart files.
We modified init.c as described in that note; rebuilt loader and put it back
in the initrd.img. Problem solved. Not clean, not pretty, but effective!
Just a hack to get round the STP timers. Maybe a cleaner option would
not be to reset the interface if booted via PXE, although how would
wouold detect this I don't know.
HTH
G
