[c-nsp] CDP interoperability

[Phil Mayers]

On 10/12/12 15:39, Aivars wrote:
>    Hi,
>
>    I thought that CDP essence is to help understand what device you
>    have at the other end of the wire no matter what. You just plug one

Well... not really. For starters, if you have a non-CDP-aware layer2 
device between 3 or more cisco devices, the will all see each other "on 
the end of the wire".

>    end of the cable into one box and the other end into another and
>    you get your CDP neighbors. Besides other side usage like in IP
>    phone communication with switches this is why anybody would use CDP.
>    Right?
>
>    Up until this morning I also thought that CDP frames are always sent
>    untagged. This is the way I would do it. Well, I was wrong. Actually
>    on Catalyst switches CDPs are sent in vlan 1. If you make some other
>    vlan native on a trunk port, CDPs are sent with dot1q tag "1". vlan
>    dot1q tag native will also do the same trick.
>
>    Now imagine a brand new shiny IOS-XR box, ASR9k for example. If it
>    has no subinterface configuration with encapsulation dot1q 1, CDP
>    will be broken. It will send CDPs with no tag and Catalyst will be
>    happy about it. It will show ASR as CDP neighbor. ASR instead
>    doesn't now what a hell tag "1" means and drop these frames.
>
>    Cisco thinks - this is expected behavior.

Because it is. You're describing very old CDP behaviour. Cisco can 
define it any way they like, because it's their protocol.

>    What do you guys think? Is this a bug or a feature? Should it remain
>    as it is?

People should stop using CDP if possible and use LLDP, which specifies 
how this case should be handled.