Signs of Triviality

Opinions, mostly my own, on the importance of being and other things.
[homepage] [index] [jschauma@netmeister.org] [@jschauma] [RSS]

Netbooting vs. spanning tree

A number of cluster nodes ('node's) are connected to an extreme networks switch ('Summit'), netbooting off a fileserver ('doppelbock') connected to the same switch. So far so good. I now have added a number of new nodes ('NNode's), but since the Summit switch is full, I connected the Summit switch to a Cisco Catalyst 4506 ('cat4506'), to which the NNodes are connected. cat4506 has three vlans.

This looks like so:


[ node ]+                                           +[ NNode ]
[ node ]+                                           +[ NNode ]
[ node ]+----- [ Summit ] ------- [ cat4506 ] ------+[ NNode ]
[ node ]+          |                                +[ NNode ]
[ node ]+          |                                +[ NNode ]
             [ doppelbock ]
    
Summit has all nodes on one vlan, cat4506 has all NNodes on one vlan, and the connection between the two switches is on the same vlan, as is the connection to doppelbock.

The nodes can netboot off doppelbock just fine, but the NNodes can't. They pxeboot and get the right IP address from doppelbock, they tftp the kernel and all's well, but when the kernel tries to DHCP, it times out. If the NNodes are connected directly to the Summit, they netboot fine.

Solution:
no spanning-tree vlan vlanid
The NNodes flutter the link signal, and the cat4506 switch will sit there not forwarding packets until it's sure you don't have a switch connected. Turning off spanning-tree on the vlan in question solves the problem.

February 22, 2006


[Mailman + https] [index] [Mac OS X NFS share vs. Garritan Personal Orchestra]