[ipxe-devel] 0x040ee119 error if heavy traffic on line

Oliver Rath rath at mglug.de
Mon Jan 20 22:14:31 UTC 2014

Hi Matt!

On 20.01.2014 22:04, Matthew Helton wrote:
> Oliver,
> Even the largest DHCP servers don't require bonded interfaces except
> in very specific configurations.
Im using dnsmasq with bond0 as listening interface. At the moment Im
running bond mode 3, not 802.1ad, but i wanted to try. this.
Alternativly i could set separate subnets for each of the NIC.

> This is probably an issue with your DHCP Server: If you are running
> your DHCP server on a Bonded interface (generally not required), there
> may be some switch-side configuration settings you may want to look
> into ('PortFast' or 'Portfast Edge'. If you are using a 802.1ad
> Interface with Load Balancing (Round Robin) "Mode 4" as the serving
> interface for DHCP, you can easily run into problems if you happen to
> have a "Mode 4" VPC configured across an A-B switch pair.

Portfast is a good idea! Im using Cisco-switches with level-3
Management, so Im sure this is settable. Im not using VPC at the moment.

> In general, I have seen more problems caused with UDP-based protocols
> and 802.1ad "Mode 4" than simple failover Modes ("Mode 1" and "2"), as
> a consequence I have shied away from using 802.1ad "Mode 4" for DHCP
> and TFTP Services.  I have had much better results using 802.1ad "Mode
> 4" on TCP-based protocols (Web and File Services).    

Yeah. My first results with mode 4 (without manageable switches) got
spurios timeouts. So i changed to mode 3 for first.

> A simple test for this would be to disconnect all but one interface
> from the server and run a couple of traces vs. all interfaces connected.

I have the following configuration in detail: Every of the 4 cards is
connected to a separate (manageable) swich, every switch is connected to
up to 9 clients. So disconnecting would result in disabling a group of

Maybe separating into subnets seems the best way, isnt it?


More information about the ipxe-devel mailing list