Broadcom NIC delays prevent DHCP during Kickstart

[Simon Gao]

We also see the similar problems with Dell and other vendor machines
using Broadcom NIC. It's not just servers, but desktop machines also. 
We use HP ProCurve switch. However, machines having Intel NIC do not
have any noticeable delay when machine comes up. The long delay seems
only happen to Broadcom NIC. As a result, some NFS share does not get
mounted.

I would be very interested in finding out how to fix the problem. I am
hesitant to make changes to backend switch since there has been no such
problem until the machines with Broadcom NIC brought onto the network.

Simon


Nicola_Worthington at mcafee.com wrote:
> I'm also seeing this problem, but only on our PowerEdge servers. -- It
> doesn't happen with out IBM NetVistas. --
>  
> We do not have spanning tree enabled. Our switching is pretty vanilla,
> with the odd vlan.
>  
> We have a DHCP master and slave which respond within 5 seconds to the
> PXE DHCP requests. When kickstart asks DHCP for an IP it *always*
> fails. We don't see the requests come in to the DHCP servers. If we
> retry, it will usually fail. We have found that sometimes unplugging
> the kickstarting machine from the network after it has PXE booted,
> leaving it a couple of minutes and then retrying, will sometimes fix
> the problem, but this isn't a propper solution. We'd much rather try
> to find out what's going wrong and get kickstart to play nicely on our
> PowerEdge servers than move to fixed IP kickstarts.
>  
> We're seeing the problem on 750s, 1850s, 1950s and 2850s.
>
>     ------------------------------------------------------------------------
>     *From:* linux-poweredge-bounces at dell.com
>     [mailto:linux-poweredge-bounces at dell.com] *On Behalf Of *Kevin Kwast
>     *Sent:* 20 October 2006 05:06
>     *To:* Mann, Andrew; Linux-PowerEdge at dell.com
>     *Subject:* Re: Broadcom NIC delays prevent DHCP during Kickstart
>
>     In my case, sure, spanning tree is on because every 5324 switch is
>     interconnected with at least one other, and I haven't used the web
>     interface to repeatedly set STP off for every port that is
>     connected to a host.  Maybe the command line offers a way to turn
>     off STP for 21 or 22 ports at a time on each switch?  I know that
>     STP can cause delays after bringing up the ethernet link on a
>     port, but Intel NICs don't seem to have this problem.  Is there a
>     way to keep the ethernet link up between when the PXE boot uses it
>     and the kickstarting Linux kernel tries to DHCP?  Or add 30
>     seconds to the amount of time it waits before it does the DHCP? 
>     Thanks!
>
>     On 10/19/06, *Mann, Andrew* <amann at ea.com <mailto:amann at ea.com>>
>     wrote:
>
>
>         > Is spanning tree enabled on the switch?
>
>           Is there a conflict with the Broadcom NICs and spanning
>         tree?  We're
>         using a Cisco 4500 series switch with spanning tree
>         enabled.  2550s,
>         1750s, 1850s, and 1955 blades are able to quickly bring the
>         NIC up and
>         start communicating.  2650s and 2950s have about a 30-60
>         second delay
>         between the time the NIC is enabled and communication
>         succeeds.  After a
>         reboot, we have to run through and restart NTP on each of
>         these types of
>         system since this delay causes the ntp startup to fail.  It's
>         only been
>         mildly annoying in the past, but getting rid of it would be
>         nice :)
>
>
>         Andrew
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Linux-PowerEdge mailing list
> Linux-PowerEdge at dell.com
> http://lists.us.dell.com/mailman/listinfo/linux-poweredge
> Please read the FAQ at http://lists.us.dell.com/faq