Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?


PIM Join Load Balancing on Multipath MVPN Routes Overview

A multicast virtual private network (MVPN) is a technology to deploy the multicast service in an existing MPLS/BGP VPN.

The two main MVPN services are:

  • Dual PIM MVPNs (also referred to as Draft-Rosen)

  • Multiprotocol BGP-based MVPNs (also referred to as next-generation)

Next-generation MVPNs constitute the next evolution after the Draft-Rosen MVPN and provide a simpler solution for administrators who want to configure multicast over Layer 3 VPNs. A Draft-Rosen MVPN uses Protocol Independent Multicast (PIM) for customer multicast (C-multicast) signaling, and a next-generation MVPN uses BGP for C-multicast signaling.

Multipath routing in an MVPN is applied to make data forwarding more robust against network failures and to minimize shared backup capacities when resilience against network failures is required.

By default, PIM join messages are sent toward a source based on the reverse path forwarding (RPF) routing table check. If there is more than one equal-cost path toward the source [S, G] or rendezvous point (RP) [*, G], then one upstream interface is used to send the join messages. The upstream path can be:

  • A single active external BGP (EBGP) path when both EBGP and internal BGP (IBGP) paths are present.

  • A single active IBGP path when there is no EBGP path present.

With the introduction of the multipath PIM join load-balancing feature, customer PIM (C-PIM) join messages are load-balanced in the following ways:

  • In the case of a Draft-Rosen MVPN, unequal EBGP and IBGP paths are utilized.

  • In the case of next-generation MVPN:

    • Available IBGP paths are utilized when no EBGP path is present.

    • Available EBGP paths are utilized when both EBGP and IBGP paths are present.

This feature is applicable to IPv4 C-PIM join messages over the Layer 3 MVPN service.

By default, a customer source (C-S) or a customer RP (C-RP) is considered remote if the active rt_entry is a secondary route and the primary route is present in a different routing instance. Such determination is being done without taking into consideration the (C-*,G) or (C-S,G) state for which the check is being performed. The multipath PIM join load-balancing feature determines if a source (or RP) is remote by taking into account the associated (C-*,G) or (C-S,G) state.

When the provider network does not have provider edge (PE) routers with the multipath PIM join load-balancing feature enabled, hash-based join load balancing is used. Although the decision to configure this feature does not impact PIM or overall system performance, network performance can be affected temporarily, if the feature is not enabled.

With hash-based join load balancing, adding new PE routers to the candidate upstream toward the C-S or C-RP results in C-PIM join messages being redistributed to new upstream paths. If the number of join messages is large, network performance is impacted because of join messages being sent to the new RPF neighbor and prune messages being sent to the old RPF neighbor. In next-generation MVPN, this results in BGP C-multicast data messages being withdrawn from old upstream paths and advertised on new upstream paths, impacting network performance.

In Figure 1, PE1 and PE2 are the upstream PE routers. Router PE1 learns route Source from EBGP and IBGP peers—the customer edge CE1 router and the PE2 router, respectively.

Figure 1: PIM Join Load BalancingPIM Join Load Balancing
  • If the PE routers run the Draft-Rosen MVPN, the PE1 router distributes C-PIM join messages between the EBGP path to the CE1 router and the IBGP path to the PE2 router. The join messages on the IBGP path are sent over a multicast tunnel interface through which the PE routers establish C-PIM adjacency with each other.

    If a PE router loses one or all EBGP paths toward the source (or RP), the C-PIM join messages that were previously using the EBGP path are moved to a multicast tunnel interface, and the RPF neighbor on the multicast tunnel interface is selected based on a hash mechanism.

    On discovering the first EBGP path toward the source (or RP), only new join messages get load-balanced across EBGP and IBGP paths, whereas the existing join messages on the multicast tunnel interface remain unaffected.

  • If the PE routers run the next-generation MVPN, the PE1 router sends C-PIM join messages directly to the CE1 router over the EBGP path. There is no C-PIM adjacency between the PE1 and PE2 routers. Router PE3 distributes the C-PIM join messages between the two IBGP paths to PE1 and PE2. The Bytewise-XOR hash algorithm is used to send the C-multicast data according to Internet draft draft-ietf-l3vpn-2547bis-mcast-bgp, BGP Encodings and Procedures for Multicast in MPLS/BGP IP VPNs.

Because the multipath PIM join load-balancing feature in a Draft-Rosen MVPN utilizes unequal EBGP and IBGP paths to the destination, loops can be created when forwarding unicast packets to the destination. To avoid or break such loops:

  • Traffic arriving from a core or master instance should not be forwarded back to the core facing interfaces.

  • A single multicast tunnel interface should either be selected as the upstream interface or the downstream interface.

  • An upstream or downstream multicast tunnel interface should point to a non-multicast tunnel interface.

As a result of the loop avoidance mechanism, join messages arriving from an EBGP path get load-balanced across EIBGP paths as expected, whereas join messages from an IBGP path are constrained to choose the EBGP path only.

In Figure 1, if the CE2 host sends unicast data traffic to the CE1 host, the PE1 router could send the multicast flow to the PE2 router over the MPLS core due to traffic load balancing. A data forwarding loop is prevented by ensuring that PE2 does not forward traffic back on the MPLS core because of the load-balancing algorithm.

In the case of C-PIM join messages, assuming that both the CE2 host and the CE3 host are interested in receiving traffic from the source (S, G), and if both PE1 and PE2 choose each other as the RPF neighbor toward the source, then a multicast tree cannot be formed completely. This feature implements mechanisms to prevent such join loops in the multicast control plane in a Draft-Rosen MVPN scenario.


Disruption of multicast traffic or creation of join loops can occur, resulting in a multicast distribution tree (MDT) not being formed properly due to one of the following reasons:

  • During a graceful Routing Engine switchover (GRES), the EIBGP path selection for C-PIM join messages can vary, because the upstream interface selection is performed again for the new Routing Engine based on the join messages it receives from the CE and PE neighbors. This can lead to disruption of multicast traffic depending on the number of join messages received and the load on the network at the time of the graceful restart. However, nonstop active routing (NSR) is not supported and has no impact on the multicast traffic in a Draft-Rosen MVPN scenario.

  • Any PE router in the provider network is running another vendor’s implementation that does not apply the same hashing algorithm implemented in this feature.

  • The multipath PIM join load-balancing feature has not been configured properly.