I'm in a debate with a guy over the use of 'ip address x.x.x.x s.s.s.s
secondary' on Cisco gear. I seem to remember reading that the use of
secondary addresses is a bad idea, but I can't recall the details of
why. Process switched? Can anyone offer a resource or more specific
information?
Thanks,
Dan
I'm in a debate with a guy over the use of 'ip address x.x.x.x s.s.s.s
secondary' on Cisco gear. I seem to remember reading that the use of
secondary addresses is a bad idea, but I can't recall the details of
why. Process switched?
No, traffic to hosts within a subnet configured as secondaries
will be CEF switched.
The only "bad" thing I can think of with secondaries, is that it's often
not what you want, why not split it on layer 2 as well, and get the
benefit of a smaller broadcast domain ?
Can anyone offer a resource or more specific information?
/Jesper
A few other possible issues:
1) routing protocols (i.e. ospf) will not form adjacencies with devices in
the secondary address subnets...so if you're doing this to get more
address space on a particular ethernet without renumbering, if you need
OSPF on the ethernet, all the OSPF speakers have to be in the primary
subnet.
2) If you're doing this to separate customers, it doesn't really. They're
all free to steal each others IPs. Better solutions would be VLAN
trunking back to the router with a subint for each subnet or a L3 switch
effectively doing that all in one box.
3) Human error. More than once I've seen someone change an interface's
primary IP by "adding a secondary" and hitting return before typing
"secondary". Maybe it would have been better/safer if the command were
"secondary ip addr ..." 
Jon Lewis wrote:
I'm in a debate with a guy over the use of 'ip address x.x.x.x s.s.s.s
secondary' on Cisco gear. I seem to remember reading that the use of
secondary addresses is a bad idea, but I can't recall the details of
why. Process switched?No, traffic to hosts within a subnet configured as secondaries
will be CEF switched.The only "bad" thing I can think of with secondaries, is that it's often
not what you want, why not split it on layer 2 as well, and get the
benefit of a smaller broadcast domain ?A few other possible issues:
1) routing protocols (i.e. ospf) will not form adjacencies with devices in
the secondary address subnets...so if you're doing this to get more
address space on a particular ethernet without renumbering, if you need
OSPF on the ethernet, all the OSPF speakers have to be in the primary
subnet.2) If you're doing this to separate customers, it doesn't really. They're
all free to steal each others IPs. Better solutions would be VLAN
trunking back to the router with a subint for each subnet or a L3 switch
effectively doing that all in one box.
I meant to add (but apparently didn't sent the reply where I thought I
did):
Depending on traffic flows, the "one-armed" routing (bouncing the
traffic from one IP net to the other off the router) can be a
significant issue for the router.
3) Human error. More than once I've seen someone change an interface's
primary IP by "adding a secondary" and hitting return before typing
"secondary". Maybe it would have been better/safer if the command were
"secondary ip addr ..."
That is an especial treat when you do it the interface you are talking
to the router on.
I always set a secondary on the most-likely-to-be-the-managment
interface and left it there and used it for managment sessions.
One which hasn't been mentioned - DHCP will break horribly if the dhcp
shared-subnets declarations don't match the multinetted subnets on the wire.
