Routing

 View Only
Back to discussions
Expand all | Collapse all

Real-world effectiveness of SR-TE + PCE for autonomous traffic engineering

  • 1.  Real-world effectiveness of SR-TE + PCE for autonomous traffic engineering

    Posted 15 days ago
    Hello everyone,
     
    I currently work in a large ISP/backbone environment where we frequently deal with traffic engineering challenges caused by fiber cuts and capacity loss events.
     
    Today, most of our traffic engineering is still handled manually using RSVP-TE/LDP based approaches. In some scenarios, we may lose hundreds of gigabits or even multiple terabits of available capacity, requiring constant tunnel adjustments and operational intervention.
     
    Because of that, I have been studying Segment Routing (SR-MPLS) combined with centralized PCE approaches and would like to hear real-world feedback from engineers who already operate this architecture in production.
     
    Some questions I would love to hear opinions about:
     
    - How effective is SR-TE + PCE in real production environments for autonomous traffic engineering?
    - Does it significantly reduce operational complexity compared to traditional RSVP-TE?
    - How well does it behave during major backbone failures and large traffic shifts?
    - Are there any scaling or operational caveats that became apparent only after deployment?
    - Have you seen measurable improvements in convergence, capacity utilization or network stability?
    - For medium/large ISP backbones, is SR-MPLS currently the preferred migration path over traditional MPLS/LDP architectures?
     
    Our environment is multi-vendor and currently heavily MPLS/LDP based, so I'm especially interested in migration experiences and operational lessons learned.
     
    Would really appreciate insights from engineers who have deployed this at scale.
     
    Thanks!


    ------------------------------
    Guilherme Contino
    ------------------------------


  • 2.  RE: Real-world effectiveness of SR-TE + PCE for autonomous traffic engineering

    Posted 10 days ago

    (please excuse the bias, as I come from the vendor side).

    Hi Guilherme,

    # if you're after a distributed approach for bandwidth management, a proven and scalable way is to use RSVP-TE with:

     - auto-bandwidth

     - container LSPs

     - on-demand next-hops (known as "dynamic-tunnels" in Junos)

     - various optimization knobs depending on the final intent 

    It is standards-based, so the vendor compatibility will be probably best.

    The requirement for this approach is that most of your traffic in the core (where capacity/bandwidth management is mostly required) is tunneled with RSVP-TE. Traffic from PEs may be still LDP or SR-based, and tunneled in the core over RSVP-TE LSPs.

    If you are looking for a mid- or longer term solution, please have a look at MPTE: https://datatracker.ietf.org/doc/html/draft-kompella-teas-mpte-02

    # if you're after a centralized approach:

    SR-TE doesn't have hop-by-hop signaling, so it cannot do bandwidth reservations and admission control. As you mentioned, you'd need a controller. This means shifting the complexity from the network elsewhere:

     - switching to a centralized model instead of distributed

     - introduction of more protocols with various levels of compatibility between vendors, various level of support from the controller and routing node side:

              - to name a few: PCEP, gNMI (SNMP?), BGP-LS, possibly netconf

     - still, you'd need most of the traffic to be SR-TE-based, otherwise TE management will not be effective. A decision must be taken if it's end-to-end SR-TE (i.e. touching all PE devices) or only in the core - possible with LDP-tunneling and/or SR shortcuts with strict-SPF.

    Both approaches are possible, and both have pros and cons. Juniper/HPE have a great, time-proven and scalable implementation of RSVP-TE for the distributed bandwidth management, but as well a controller solution for a centralized approach - Juniper Routing Director, that can cover well use cases like congestion avoidance, tunnel optimizations etc.

    Thanks

     Anton



    ------------------------------
    Anton Elita
    ------------------------------