Unidirectional traffic in one way over multiple physical links

ThomasD86 · ‎01-19-2024

Hi,
I am working with the following topology:

Between PE#1 and PE#2 an iBGP neighborship exists, both PEs have an OSPF neighborship with the P router. Recently, we've added a second physical link between P and PE#2 but we observed an uneven balancing of the traffic after this operation.

Traffic from PE#1 to PE#2 that transits over P1, will only always utilize the first link (.86->.87) whereas traffic from PE#2 to PE#1 will be balanced across all 3 links.

Looking at the P1 router, I see that the route to PE#2 is known from both of the links:

Routing entry for 172.16.0.161/32
  Known via "ospf 1", distance 110, metric 26, type intra area
  Installed Jan 11 01:09:18.068 for 1w1d
  Routing Descriptor Blocks
    172.16.2.87, from 172.16.0.161, via FortyGigE0/0/0/13
      Route metric is 26
    172.16.2.201, from 172.16.0.161, via FortyGigE0/0/0/14
      Route metric is 26

However, looking at the mpls forwarding table:

show mpls forwarding prefix 172.16.0.161/32
Fri Jan 19 15:57:12.555 CET
Local  Outgoing    Prefix             Outgoing     Next Hop        Bytes       
Label  Label       or ID              Interface                    Switched    
------ ----------- ------------------ ------------ --------------- ------------
24324  Pop         172.16.0.161/32    Fo0/0/0/13   172.16.2.87     13079621827

and the CEF:

show cef 172.16.0.161
Fri Jan 19 15:57:48.195 CET
172.16.0.161/32, version 49063, internal 0x1000001 0x30 (ptr 0xa1285d40) [1], 0x0 (0x90919418), 0xa20 (0xa15b6ef8)
 Updated Jan 11 01:09:18.077 
 local adjacency to FortyGigE0/0/0/13

 Prefix Len 32, traffic index 0, precedence n/a, priority 3
   via 172.16.2.87/32, FortyGigE0/0/0/13, 3 dependencies, weight 0, class 0 [flags 0x0]
    path-idx 0 NHID 0x0 [0xa0d7a6a8 0x0]
    next hop 172.16.2.87/32
    local adjacency
     local label 24324      labels imposed {ImplNull}
   via 172.16.2.201/32, FortyGigE0/0/0/14, 6 dependencies, weight 0, class 0 [flags 0x0]
    path-idx 1 NHID 0x0 [0xa0d7bc08 0x0]
    next hop 172.16.2.201/32
    local adjacency
     local label 24324      labels imposed {None}

The first thing I notice is that if for the link over Fo0/0/0/13 PE#2 passed to the P router an impnull label, it did not do the same for the link over Fo0/0/0/14.

Looking at PE#2:

show mpls forwarding-table 172.16.0.163    
Local      Outgoing   Prefix           Bytes Label   Outgoing   Next Hop    
Label      Label      or Tunnel Id     Switched      interface              
2531       24361      172.16.0.163/32  0             Fo0/2/8    172.16.2.200
           24361      172.16.0.163/32  0             Fo0/2/0    172.16.2.86 
           24359      172.16.0.163/32  0             Fo0/2/4    172.16.2.88

Still on PE#2, looking at the forwarding-table towards the P1 router:

show mpls forwarding-table 172.16.0.20 
Local      Outgoing   Prefix           Bytes Label   Outgoing   Next Hop    
Label      Label      or Tunnel Id     Switched      interface              
230        Pop Label  172.16.0.20/32   54268855      Fo0/2/0    172.16.2.86 
           Pop Label  172.16.0.20/32   6332583       Fo0/2/8    172.16.2.200

So it seems that PE#2 is passing the correct label however it seems that this is not reflected on the P1 router. What could be the cause of this?

Thanks for helping

MHM Cisco World · ‎01-19-2024

you need to check the routing table in both P and PE
one have equal cost multi path other not have
I think this issue is mismatch of cost in both end

filopeter · ‎01-22-2024

Hi,

this is an expected behaviour on P1 router. It is an MPLS P (LSR) router, it just swaps labels according the mpls forwarding table. There must be a clear statement what to do with the incoming labels in the lfib (mpls forwarding table).

To achieve load-balancing consider to use either different bgp next-hop per l3vpn on PE2, or FAT labels between P1 and PE2 in case of l2vpn service.