Problem with STP and PXE boot installation of a Linux bridge connected to procurve switches

[Reitenbach]

Hi,


I have two 6400 switches, that are the redundant backbone, then I have two 3400
with each having a connection to each 6400 switch, to have redundant routes.

then I have a bunch of Linux servers connected to the 3400 switches. the servers
have two network interfaces, the first connected to one, the second to the
other 3400 switch. The interfaces in the server are configured as a bridge,
using the linux bridge-utils.

RSTP is enabled on all four switches.

the DHCP Servers is located in the same subnet as the servers are.

Now the problem:
when I startup a connected server first, then start the installation via autoyast,
everything works well. it takes about 10-20 seconds, to get the PXE boot working,
and then later, when the linux kernel takes over. it gets its dhcp address.
The second reboot, including dhcp, and finishing the installation works well.

the third reboot activates the bridge on the two network interface. Booting
the server with the bridge takes a while, but works.

The problem now is, when I now reboot the server, and trying to make another
PXE boot installation, then it takes about 30seconds or up to a minute, to
get the PXE boot information, later when the linux kernel takes over, and tries
a dhcp request on its own, the dhcp request times out, and the installation
stops.

There are two solutions to the problem,
1. shutdown the node, and wait a not too short time, (cannot really tell what is the
shortest amount of time to wait), then start the machine and begin PXE boot
2. reconfigure sth. in the switch, that is spanning tree related, e.g. enable/disable
mcheck, enable/disable edge-port, enable/disable point-to-point-mac, change the
hello-time or forward-delay, ... it does not matter, what to change, but when I
reboot the system after such a change, the PXE boot and the installation has no
problem to get the needed information from the dhcp server


now my questions:
1. Anybody can explain me, or point me to some documentation, why this above described
situation happens?
2. can I configure the switches, to be able to reboot the servers and start an installation
without the need to reconfigure the switch or wait a long time before doing so?

kind regards
Sebastian

[Richard Brodie_1]

I'm not enough of a Linux guru to understand why you are running the bridging software on them. However, I would imagine what is happening is that it is forcing the HP switches to put their ports into plain old STP mode (I don't believe bridge-utils supports RSTP).

[Matt Hobbs · Accepted solution]

I agree with Richard, it sounds very much like bridge-utils only supports STP at the moment... I did find this though that indicates it will support 802.1w (RSTP):

http://linux-net.osdl.org/index.php/Bridge#Are_there_plans_for_RSTP_.28802.1w.29_support.3F

What NICs are you using? I would look into the proper teaming driver/application for whatever NICs you are using. It'll give you load balancing and won't be affected by this spanning-tree problem.

[Reitenbach]

Thanks a lot, both of you. After resarching a bit, it must be as you explained it, it is just that the linux hosts do not make RSTP, and it just takes a while for the switch to "switch back" to RSTP after it was in STP mode.

[Reitenbach]

close