vMX

last person joined: 5 days ago 

Ask questions and share experiences about vMX.

  • 1.  Load balancing over LSP on MX

    Posted 06-06-2020 22:03

    Hi everybody,

    I am trying to understand how traffic can be load balanced over two LSP .

    I stumbled upon this Juniper KB:

    https://kb.juniper.net/InfoCenter/index?page=content&id=KB12995&actp=METADATA

    Capture-TEST.PNG

     

    As per link, following is configured on ingress Chewie:

    forwarding-options {
    hash-key {
        family mpls {
            label-1;
            label-2;
            payload {
                ip;

     

    lab@Chewie> show route forwarding-table destination 10.1.1.3 
Routing table: inet
Internet:
Destination        Type RtRef Next hop           Type Index NhRef Netif
10.1.1.3/32        user     0                    ulst 262143     1
                                                Push 299824       t1-2/0/1.0
                                                Push 299808       t1-2/0/0.0

    Above we see Chewie will load balance traffic destined to 10.1.1.3/32 over two LSP.

     

    Are we saying:

    1) if Chewie ( ingress LSR) receives MPLS labled packets , it will use MPLS lable1 , label2 and ip load to compute hash key for load balacing?

    If above is true, how about traffic which is not labelled ?  for example, Igress LSR ( Chewie) issues pings to 10.1.1.3 , which is not labelled yet , how Chewie load balance traffic over these two LSP as there is no MPLS lablels but our load balancing configuration , as shown in forwarding option at the start of this post,  says please factor MPLS label1 , label2 and IP pay load to compute hash?

     

    Thanks and have a good night!!

     

     



  • 2.  RE: Load balancing over LSP on MX
    Best Answer

    Posted 06-06-2020 22:32

    Hello,

    You found a VERY OLD article 🙂

     

    Load-balancing traffic over RSVP-based MPLS Label-switched paths (LSPs)

     
     
    Point by point:
     
    1/ "hash-key" on MX is applicable only to DPC linecards and has no effect for MPC linecards.
    MPC linecards use "enhanced-hash-key" which has L3 and L4 loadbalancing enabled by default.
     
    2/ 
    @sarahr202 wrote:

     

    Are we saying:

    1) if Chewie ( ingress LSR) receives MPLS labled packets , it will use MPLS lable1 , label2 and ip load to compute hash key for load balacing?


    If Chewie received LABELED MPLS packets, then it is NOT an ingress LSR, it is simply LSR.
    But to answer Your question - if transit Chewie LSR receives LABELED MPLS packets because transit MPLS LSP passes thru Chewie, then Chewie will forward these packets along that single transit LSP. No LSP load-balancing happens on transit LSR - simply because p2p MPLS LSP does not split/branch out. P2mp MPLS LSP also do not load-balance, they replicate.
    If You are interested in the MPLS LSP load-balancing on transit LSR over EGRESS LAG/Ether-channel then yes, the balancing will happen according to configured "hash-key" for DPC cards and "enhanced-hash-key" for MPC cards.
     
     
     3/ 
     
    @sarahr202 wrote:

     

    how about traffic which is not labelled ?  for example, Igress LSR ( Chewie) issues pings to 10.1.1.3 , which is not labelled yet , how Chewie load balance traffic over these two LSP as there is no MPLS lablels but our load balancing configuration , as shown in forwarding option at the start of this post,  says please factor MPLS label1 , label2 and IP pay load to compute hash?

     


    There are 2 separate questions here.
    3a/ the locally-generated unlabeled traffic - pings FROM Chewie - will NOT be load-balanced. This is by design and You cannot influence this.
    3b/ the unlabeled TRANSIT traffic will be load-balanced across 2 LSPs according to configured "hash-key"/"enhanced-hash-key".
     
    HTH
    Thx
    Alex