[j-nsp] l2circuit between ASR9k and MX80

[Saku Ytti]

On (2015-08-04 07:57 +0200), Marcin Kurek wrote:

Hey Marcin,

> Actually I'm using 2nd gen ASR9k line cards, that is Typhoon. Just to make
> it clear, did you test this on CAT7600 ES+ on one side, and JNPR on the
> other?

Yes.

> Honestly I'm a bit lost here. ASR9k allows me to manipulate only 2 topmost
> tags.
> Are you saying that if I have a frame coming to PE:
> [ SVLAN2 SVLAN1 CVLAN1 ]
> I should get rid of SVLAN2, then at the far-end swap SVLAN1 and push back
> SVLAN2, whatever it needs to be?

Preferably you'd get rid of SVLAN1, instead of SVLAN2, as SVLAN2 is more
likely to have relevant CoS bits.
But yes, always standardize incoming frames to 1 SVLAN, then you don't have to
know B side config, to provision A side.

If you have no SVLANs, then you'll push one towards MPLS core and pop when it
comes from MPLS core.

> I didn't try this on MX yet, but it seems that on ASR9k I cannot perform
> multiple operations at once, like:

You don't need translate on Cisco, it does it automatically, by magic for last
10 years (the original EoMPLS version didn't do it).

> So if I'm dealing with a scenario where I'm having a single SVLAN:
> 
> [SVLAN1 CVLAN]
> 
> what is the real difference between the two cases?
> 
> a) popping SVLAN1 and pushing it back on the other side
> 
> ----A----
> interface GigabitEthernet0/1/0/19.100 l2transport
>  encapsulation dot1q 100
>  rewrite ingress tag pop 1 symmetric
> 
> 
> ----B----
> interface GigabitEthernet0/1/0/19.200 l2transport
>  encapsulation dot1q 200
>  rewrite ingress tag pop 1 symmetric
> 
> b) leaving the stack as-is on ingress and swapping SVLAN1 on the other side

As per cisco, it does not actually send it the the core without VLAN tag, as
it's VLAN mode EomPLS, it really does add some cruft back.
So yes, this would end up transporting the SVLAN and be an acceptable config.

> ----A----
> interface GigabitEthernet0/1/0/19.100 l2transport
>  encapsulation dot1q 100
> 
> 
> ----B----
> interface GigabitEthernet0/1/0/19.200 l2transport
>  encapsulation dot1q 200
>  rewrite ingress tag translate 1-to-1 dot1q 100 symmetric

This looks problematic, not sure if it even works, but if it does, B needs to
know A side config when provisioning, this type of dependency should be
avoided when possible.

> Is it a matter of retaining CoS marking?

For me that is important, that is why I've standardized for 1 SVLAN. If
retaining CoS is irrelevant, then you could standardize to 0 SVLAN. But I'd
recommend against that. 4 byte is small price to pay, for avoiding potentially
lot of work in future.

-- 
  ++ytti