FEC Label Binding Procedures for Flows to the Same Egress Router

[ShahriarBasiri]

Dear Community,

According to RFC3036:

  If it is known that a packet must traverse a particular egress
  router, and there is an LSP which has an Address Prefix FEC
  element which is an address of that router, then the packet is
  mapped to that LSP.  The procedure for obtaining this knowledge
  is beyond the scope of this document. (As an example, if one is
  running a link state routing algorithm, it may be possible to 
  obtain this information from the link state data base. As another
  example, if one is running BGP, it may be possible to obtain this
  information from the BGP next hop attribute of the packet's route.)

 If such binding can be possible somehow, there will be at most ( num_of_ingress_routers × num_of_ingress_routers × num_of_traffic_classes) labels needed. In such a case, there will be a great decrease in number of labels in comparison to assigning one label for every prefix in routing table. Here is my questions. I will really appreciate it if you would answer any of these:

1- Can you explain the examples more? for the second example I guess it means that because MPLS is usually inside a single area (usually backbone area) and because AS border router is usually an edge router, routers inside the MPLS network can use the same label for all traffic destined for that ASBR. Is it right? Still I can not understand first example and how we can find out from link state data base the identity of egress router.

2- Is there any other procedures for identifying flows which traverse the same path to the same egress routers?

3- Is any of these procedures implemented in Cisco?

4- This question is a bit irrelevant to above questions. Is there a way to enable MPLS just for BGP traffic only (traffic whose destination is outside of an autonomous system) in Cisco routers?

Any help is really appreciated, 

Shahriar.

[Giuseppe Larosa]

Hello @ShahriarBasiri ,

The RFC 3306 refers to LDP operations.

We usually confgure routers in the following manner:

OSPF RID = LDP RID = BGP RID = loopback IP address

LDP builds unidirectional LSPs to each possible LSR.

The way that LDP build LSPs is using node wide label space, each device performs label bindings for the loopbacks of remote devices.

you can see an LSP as built from destination to local router by the fact that:

penultimate hop receives an implicit null IPv4 label = 3

each of the upstream routers adverttise its local label to the other routers.

So the actual label value in an LSP with destination a remote loopback IP address changes its value at each router hop.

However, there is recursion : BGP prefixes in AF IPv4 unicast don't need a specific label for each of them because the LSP used to reach the BGP next-hop can be used for BGP prefixes too.

To save on MPLS label consumption most ISP implement label filtering both in local allocation and in propagation.

if all loopbacks are /32 and they are taken from a common address block like 10.255.250.0/23 we need just to create LSPs for those destinations.

In this way we avoid to waste the limited label space for prefixes like inter router links that do not provide any advantage.

>> 2- Is there any other procedures for identifying flows which traverse the same path to the same egress routers?

LDP uses the IGP information and BGP can use the concept of recursive routing on the BGP next-hop to resolve BGP prefixes on the LSP with destination = BGP next-hop.

>> 3- Is any of these procedures implemented in Cisco?

MPLS label filtering for allocation of local labels and for advertsing is a good example of label use reduction.

>>> 4- This question is a bit irrelevant to above questions. Is there a way to enable MPLS just for BGP traffic only (traffic whose destination is outside of an autonomous system) in Cisco routers?

Yes by using MPLS label filtering BGP prefixes can use the LSPs to BGP next-hops as explained above but the next-hop has to be known in IGP so usually the border routers having eBGP sessions will use next-hop self towards iBGP peers.

So these two tricks MPLS label filtering used to advertise only loopbacks in LDP and next-hop self work together.

To be noed wihtout the LSPs for the BGP next-hop it is not possible to build a working MPLS forwarding plane, because the BGP next-hop is checked first.

Hope to help

Giuseppe