ASR9K not removing MPLS label on Virtual Interfaces (BVI)

[Jacobo Schneider]

We are facing issues with the MPLS Label removal and Label generation when the exit interface from the MPLS domain is a logical (not physical) interface.

[xthuijs]

Yeah, running mpls is not supported over BVI.

That is LDP cant run over BVI, nor can the BVI subnet be advertised as a per-prefix label.

You will want to send the BVI network as an aggregate label at the minimum so that it will become a 2 stage lookup for the ingress card.

the ingress linecard doesnt have enough context available to do per-prefix on a BVI (since a BVi can serve multiple L2 interfaces on different linecards).

For that reason we want the aggregate label so that we get one pass through the NPU on ingress to pop the label, and the next pass to do effectively simple plain BVI L3 to L2 forwarding.

See here for the BVI restrictions (note that with XR43 we can apply QOS, no shaping, marking nad policing only, and ACL's to the BVI, which then will get replicated to the EFP's.

http://www.cisco.com/en/US/docs/routers/asr9000/software/asr9k_r4.1/interfaces/configuration/guide/hc41irb.html#wp1011723

xander

[Marks Maslovs]

Hi Alex!

I would like to ask, has there something changed for running mpls over BVI interfaces for rel. 5.1.0? According to doc, there are no difference, http://www.cisco.com/en/US/docs/routers/asr9000/software/asr9k_r5.1/interfaces/configuration/guide/hc51irb.html#wp1011723

But after the upgrade from 4.2.3 to 5.1.0 we have encountered problems with connectivity to BVI interfaces, despite that config remained unchanged (with per-vrf label allocation )

router bgp 999999

vrf rnc

  rd 1:42

  address-family ipv4 unicast

   label mode per-vrf

Thank You in advance!

[xthuijs]

Marks: in XR510 there is some support added for MPLS over BVI.

but you're having the BVI in a vrf I would assume here and there is no LDP running on the BVI is that correct?

This is for typhoon cards only pretty much, so I think there may be a tricky situation going on.

Few things to try:

-try per prefix allocation

-send some traffic at a high rate and determine with np counters where it gets dropped and for what reason that might give us a pointer.

Also one important piece is, is this trident based linecards or typhoon?

regards

xander

[Marks Maslovs]

Thanks Alex! Yes, there is no ldp running on the bvi int.

These are trindent based lc. The topology is following c7613-asr, and asr has bvi interfaces. Between c7613 and asr mpls is running. When I am pinging bvi int from c7613 in the vrf, asr drops the requests. According to routing table, all paths are known...

I've tried per prefix allocation. The result is the same.

What relates to high rate traffic, I have pinged with few thousands (3333) requests and got following results:

I see that on the np

PARSE_FAB_INJECT_MPLS_CNT

RESOLVE_AGE_MAC_STATIC_DROP_CNT

These counters are changing accordingly to number of pings sent..
yeah, should have checked this one before writing, as I understand that's not related to mpls..

[adorins]

I have the same "problem" that subnets which should be accessible through BVI interface can not be accessed if packet arrives from MPLS cloud (through mpls enabled interface). BVI is configured under vrf, and it has a static route which is then redistributed in bgp:

interface BVI6
 description test6
 vrf test6
 ipv4 address 192.168.0.1 255.255.255.0

router static
 vrf test6
  address-family ipv4 unicast
   192.168.4.0/24 192.168.0.2

Router is asr9006 with xr5.3.4 and both interfaces (mpls and exit i/f in bridge-group) is on Trident card. Label allocation mode is default which is per-prefix I believe.

I know that problem is in label allocation mode for BVI vrf (should be changed to per-vrf) but I found out a workaround which makes things to run if I modify a static route like this:

router static
 vrf test6
  address-family ipv4 unicast
   192.168.4.0/24 BVI6 192.168.0.2

May be you have any comment on this? Can this w/a be used in production without any side effects?

br

Agris

[Aleksandar Vidakovic]

hi Agris,

the combination of MPLS and BVI is not supported on Trident. Please refer to

http://www.cisco.com/c/en/us/td/docs/routers/asr9000/software/asr9k_r5-3/interfaces/configuration/guide2/b-interfaces-cg53x-asr9k/Configuring_Integrated_Routing_and_Bridging.html

"Multi protocol Label Switching (MPLS) on BVI is supported on 
Cisco ASR 9000 Enhanced Ethernet Line Cards but not on Cisco ASR 9000 Ethernet Line Cards."

 /Aleksandar

[adorins]

There is no mpls configured on bvi interface. It is just a regular ipv4 interface under vrf.

[xthuijs]

your workaround is very neat, what you are achieving is a FEC with a summary, that is what you ideally need for the forwarding to right.

a per-prefix label is a direct forwarding adjancency and that is tricky with bvi since we can have multiple EFP's in that bridge domain that can reside on different cards so that relation between prefix and outoging interface is not 1 to 1, unless the tx adj is taken into consideration.

So with per-vrf or your route trick, we create an agg label which means that we first strip the label to find the prefix and than do a regular route lookup. that is what is needed for the bvi forwarding when it is in a vpn.

xander

[Aleksandar Vidakovic]

That is correct. However, when the MPLS packet is received from the core, the next-hop interface is BVI. This is why I was referring to a combination of MPLS and BVI.

[adorins]

Thanks, guys. That is what I asked for. I read about MPLS+BVI+VRF thing and just wondered if "problem" can be solved like I did it with static route. It turns out that I did the same what can be done by changing a label allocation mode, just in different way.

[Aleksandar Vidakovic]

yes, changing the label allocation to per-vrf will make it work.