Q-in-Q for a specific VLAN on trunk interface

[Nightheaven]

Greetings all. I was wondering if there are any options to make my idea possible or if there is an alternative solution for it.

So, I need to make a Q-in-Q tunnel between host and cisco c3750. Host connected to Cisco c6504 and at c3750 side i have an interface in a trunking mode. On c6504 i have this configuration for an interface:

c6504:

c6504#conf t
c6504(config)#int gi4/15
c6504(config-if)#switchport
c6504(config-if)#switchport mode dot1q-tunnel
c6504(config-if)#switchport access vlan 10

And my idea for c3750 was to make a subinterface for vlan 10 and make it dot1q-tunnel aswell. However, for subinterfaces you can only make them if you split interface for ALL vlans on it for this to work, i would assume, the second issue - is that c3750 does not support subinterfaces as there is this vlan interfaces feature that you would user for a similar effect. And so you can't use dot1q-tunnel options there. I can use it on a Nexus 9k that is connected to this c3750, but I would still need to split port-channel in all the usable vlan subinterfaces... 

The only other somewhat reasonable idea that is cooking in my head is to make some kind of loop on c3750 with 2 interfaces to have traffic from vlan 10 on port-channel 1 (that goes to nexus) go to interface gi1/0/47 that is connected to gi1/0/48 that has dot1q-tunnel enabled...

Any suggestions on this part would be appreciated.

[Ramblin Tech]

Not sure I understand what exactly you are trying to do (I am more of visual type; diagrams help me immensely), but... using a loopback cable between ports to handle extra encaps/decaps is not as crazy (or uncommon) as it sounds. Loopback cables provide an external means for traffic on older switches/routers to make another pass through the forwarding ASIC for extra processing. Nowadays, pipelined NPUs do this sort of thing internally via "recirculation" (sending a packet back to the beginning of the pipeline to make another pass). I also suspect that what you trying to do might be done more elegantly on gear that supports the EFP configuration model, rather than switchports, but that is not what you currently have at your disposal.

[Giuseppe Larosa]

Hello @Nightheaven ,

802.1 Q in Q is at OSI layer 2 subinterfaces have L3 meaning so they would not serve for this scope.

On some platforms like ME 3400 ME3600 I think there is support for selective Q in Q . But this selective means that you can define what VLans are in the inner 802.1Q to avoid to pass all of the VLANs known on the switch.

Hope to help

Giuseppe

[Nightheaven]

Here is a simple diagram for a topoligy that takes part in this configuration.

I need to link ESXi host to the colo2 segment. Both colo1 and colo2 has their own topologies, with a usage of their own vlan pools quite alot  of which would overlap, that is why i was going for a Q-in-Q tunnel for host  to link it to the colo2 segment via colo1 one. And as it is more of a management host, it should be fine to make a link on c3750 as it should not have that much of a traffic itself. Howeve i feel like it would be better overall and let you have a more of a scalabe topology if there would be a need for a similar thing in future or if we'll need to add storage on a host side, if I'll make a split on a Po2 on a Nexus 9k side with subinterfaces as it would make it much easier to add some more Q-in-Q tunnels and such.

Unfortunately this is all in production and I feel like doing Po split would be impossible without maintanace with downtime.

[Ramblin Tech]

I am not so sure about splitting out some of the links in the LAG bundle, but the loopback cable should work without any topology changes to the Nexus or Cat6500 switches. I did a quick PoC using CML and interface configs are quite simple.