Multipath Traffic Engineering for Segment Routing

1.1. Requirements Language

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶

The BCP 14 keywords used in this document describe the expected behavior of the centralized controller and the Junction nodes when realizing an MPTE DAG using SR Policy and Binding SIDs. They are not intended to introduce new requirements on the underlying SR Policy, PCEP, BGP, BGP-LS, or NETCONF protocols themselves.¶

4.1. MPTED

The MPTED is managed by a centralized controller, such as a Path Computation Element (PCE) acting as the MC. Topology discovery is performed using BGP-LS [RFC9552], while transport control plane signaling is achieved through controller-oriented protocols such as Path Computation Element Protocol (PCEP) [RFC5440], [RFC8231] and BGP/BGP-LS [I-D.draft-ietf-idr-segment-routing-te-policy], [I-D.draft-ietf-idr-bgp-ls-sr-policy]. The MC computes, manages, and distributes all forwarding information to the nodes participating in the MPTE DAG, which form the MPTED.¶

[I-D.draft-kompella-teas-mpte] specifies that a node in the MPTED is identified by its IPv6 loopback address. However, this document allows the use of a 32-bit dotted quad router ID as an alternative to support IPv4 or dual stack networks. This value represents the headend address of a node participating in the DAG.¶

As per [I-D.draft-kompella-teas-mpte], the controller acting as the MC is responsible for assigning the MID and incrementing the MPTED unique ID version.¶

4.2. Junction Segment

The concept of a Junction Segment is introduced to describe the signaling and forwarding behavior of a Junction node in an SR network:¶

A junction segment is:¶

• realized using SR Policy construct with a single Candidate Path, a Binding SID [RFC9256] and one or more segment lists to one or more downstream nodes.¶

• similar to a Replication Segment [RFC9524], but performs forwarding based on load balancing rather than replication.¶

• installed on nodes identified as Junction Nodes, as defined in [I-D.draft-kompella-teas-mpte].¶

• combined with one or more Junction segments across a segment routed topology domain to form an MPTE DAG tunnel.¶

Since a Junction Segment may egress to multiple downstream nodes, the endpoint of the corresponding SR Policy can be set to the null value (0.0.0.0 for IPv4, :: for IPv6). Therefore, a Junction segment is identified by its <headend, color> attribute.¶

The color value is mapped to the <MID, Version> tuple and is tracked by the controller.¶

It is RECOMMENDED that a globally consistent color value be used across all Junction Segments that belong to a single DAG instance or that serve a common DAG intent.¶

The MPTE tunnel is used by one or many SR Policies instantiated on one or many ingress nodes.¶

A Junction Segment may be used by multiple ingress SR Policies, provided that the subgraph it forwards over is fully shared by all SR Policies using it, including the set of their respective egress nodes.¶

The ingress SR Policy is responsible for initiating traffic steering into the DAG and is associated with a color value representing service intent. The junction segment, on the other hand, is a DAG forwarding construct implemented as an SR Policy with a BSID.¶

To support global optimization with make-before-break (MBB) operations across a set of ingress SR Policies, the color value used by Junction Segments SHOULD differ from the color values of the ingress SR Policies that are consumers of the DAG. This distinction allows ingress SR Policies to independently manage service steering while enabling consistent forwarding behavior across the DAG.¶

The controller is responsible for maintaining and tracking the association and semantic meaning of color values across all Junction Segments that participate in the DAG.¶

Accordingly, an implementation SHOULD treat ingress SR Policies and Junction Segments as decoupled constructs, each with their own versioning and lifecycle.¶

Since SR-based networks support specifying multiple egress interfaces using adjacency-SID sets and Node SIDs, the Junction Segment MAY include a SID list entry that identifies multiple outgoing interfaces. In addition to egress interfaces, [I-D.draft-kompella-teas-mpte] describes signaling ingress interfaces. The use of a Junction Segment omits the need for per-interface ingress signaling as a single Binding Segment attached to an SR Policy is used. All upstream originated traffic sent to a downstream Junction Node uses the same, single Junction Segment value which is a Binding Segment.¶

5.1. General

A path computation request or tunnel delegation notification is sent to the controller, specifying one or more ingress and egress nodes, along with constraints or service level agreement policy information. This request may originate from an ingress router in the network or be provisioned directly via an API to the controller. This tunnel computation request or delegation pertains to an instance of an MPTE Tunnel achieved with the ingress SR Policy.¶

The controller computes the DAG for ingress and all egress endpoints to determine all Junction nodes in the DAG to be used for the tunnel.¶

The controller signals the Junction Segment(s) to all downstream nodes, starting with the penultimate egress hop node(s) and working upwards toward the ingress nodes. The controller may perform deployments in parallel provided there is no interdependency of values between junction segments, and the controller does not update the ingress until all deployments are deemed successful.¶

If a Junction Segment cannot be deployed to a node, the controller SHOULD re-attempt the deployment up to a configurably defined maximum number of retries.¶

Depending on local policy and implementation constraints, the controller MAY execute a full rollback of the configuration, or it MAY allow the partial deployment to proceed, provided that the TE constraints continue to be satisfied.¶

Junction Segments are deployed as a unicast SR Policy with a single Candidate Path using protocols such as PCEP, BGP, or NETCONF.¶

A BSID MUST be explicitly requested or signaled to the Junction node for assignment. If the controller opts for local node assigned value, it MUST wait to signal upstream Junction nodes about their Junction segments in their outgoing SID lists. The BSID value MAY be a constant value globally, if assigned by PCE/Controller, or may be a different value on each Junction Node whether it’s assigned by the controller or the local Junction Node itself.¶

Each SR Policy contains one or more SID lists. With the exception of the penultimate node, these SID lists include at least two segment identifiers: one SID for forwarding to the downstream Junction node and one for the Junction Segment value (BSID) of the downstream node. For directly connected neighbors, this may be the adjacency or node SID for the neighbor. For Junction nodes that are not directly connected, additional SIDs MAY be used to steer the packet along an ECMP or non-ECMP path to the downstream Junction node.¶

Appendix A provides an example topology and resulting Junction segments.¶

5.2. Tunnel with multiple ingress/egress

[I-D.draft-kompella-teas-mpte] specifies that an MPTE Tunnel could have multiple ingress and/or multiple egress nodes. For controller-initiated tunnels, the intended ingress and egress node(s) can be provided to the controller based on implementation specific methods. These may be signaled to the network as multiple tunnels to support multi-ingress scenarios. Each tunnel can use a null Endpoint value to support multi-egress.¶

MPTE SR Policies that are originated or defined by network devices are typically limited to a single ingress and a single egress endpoint unless protocols such as PCEP or NETCONF are extended to encode additional intended destination node(s) for controller-based path computation.¶

As mentioned in Section 4.2, the DAG tunnel may be re-used by multiple ingress SR Policies. This mechanism is used to support achieving multiple ingress nodes originated from the network, by way of the controller binding and attaching the ingress SR Policies to a pre-existing DAG sharing the same intent and endpoints.¶

5.3. Load-balancing

When a packet with the BSID assigned to the Junction Segment is received at its Junction Node, the node performs weighted non-equal cost flow-based forwarding across all egress SID lists associated with the Junction Segment.¶

As per [RFC9256] Section 2.11, The fraction of the flows associated with a given segment list is w/Sw, where w is the weight of the segment list and Sw is the sum of the weights of the segment lists. To exclude forwarding via a specific egress interface while preserving the forwarding structure, a weight value of zero is assigned to the corresponding segment list.¶

5.4. Protection

As described in [I-D.draft-kompella-teas-mpte], as there are multiple egress interfaces (SID Lists), the loss of one interface link does not result in traffic drops, as long as one egress interface (SID List) remains although congestion may occur. For example, in the topology shown in Appendix A, Node C can tolerate the loss of up to 3 egress links and traffic will still forward (potentially with congestion).¶

If a Junction Node experiences a failure of all egress links (including any protection for those links), it will initially blackhole traffic until upstream nodes are notified by the controller to remove the failed Junction node from the DAG.¶

To reduce the risk of outages caused by single link failure, the controller MAY optimize DAG deployment by assigning Junction Segments only to nodes with more than one egress segment list. In other words, if a node in the DAG has only one egress interface, it functions solely as a transit node for an upstream Junction Segment and does not receive a Junction Segment itself. For example, in the topology shown in Appendix A, Node E is omitted from receiving a Junction Segment as it only has 1 egress link.¶

Link protection from an upstream Junction node to its downstream Junction nodes can be achieved using existing Topology Independent Loop-Free Alternative (TI-LFA) [RFC9855] mechanisms, applied per egress SID List on each Junction Segment. Since the top SID(s) in each SID List identify the path to the next downstream Junction node, TI-LFA is applicable. Effectively, TI-LFA is used to protect traffic between Junction Segments along a path within the DAG and is not intended to protect traffic directed toward the DAG’s egress nodes or the entire DAG.¶

Local computation for node protection on an upstream Junction node is not feasible, because it lacks visibility into the DAG beyond the immediate downstream Junction node as it only knows the next Junction Segment. A controller MAY be used to precompute backup SR Paths and signal these backup SID Lists to the upstream Junction segments.¶

5.5. Hierarchy

The use of Junction Segments to achieve a DAG can be used in hierarchical organization and sharing of DAGs between end-to-end tunnels.¶

For instance, in a multi-area or multi-instance topology, one or more shared DAGs may be created per area, connecting the border ingress node(s) and egress node(s). These DAGs can then be stitched together for use in an end-to-end SR tunnel.¶

Figure 1 is an example of two independent SR Policy Tunnels from Headend A and Headend B terminating on Egress X. An instance of a DAG with MID 100 can be configured between area border routers (ABR) ABR-1 and the Egress X node. ABR-1 Junction Segment which ingresses the DAG has a Binding Segment attached, for example BSID-100. Therefore, the SID list on Headend A and Headend B SID lists would contain the SR Path (ex: Node-SID-ABR-1) to reach ABR 1 followed by BSID-100. The instruction set from each Headend to ABR 1 could also be another instance of a DAG, either for independent use or shared.¶

          -------------------------------------------------------------
          |                             |                             |
          |                             |                             |
     +-------+                          |             ---\            |
     |       |--\     Path 1            |         ---/  | ---\        |
     |Headend|   ----\                  |     ---/   -----\   --\     |
     |   A   |         ----\         +-----+-/   ---/      ---\  --+------+
     +-------+              ----\    |     |  --/   -------|   --  |      |
          |                      --- | ABR |  -------------------  |Egress|
          |                      --- |  1  |        |DAG-100       |  X   |
     +-------+              ----/    +-----+ --\    |        ----- +------+
     |       |        -----/            |  -\   ----\-------/     --  |
     |Headend|   ----/                  |    --\     ---       --/    |
     |   B   |--/     Path 2            |       -\   |      --/       |
     +-------+                          |         --\    --/          |
          |                             |            ---/             |
          |                             |                             |
          -------------------------------------------------------------
                   Domain/Area 1                  Domain/Area 2

                             Figure 1 : Reusable DAG 100

5.6. Directly connected Junction nodes

As described in the Operational section, all transit nodes in a DAG MAY be signaled with a Junction Segment. Alternatively depending on the topological graph and TE requirements, two Junction segments may be interconnected via an SR Path, with a SID list, where the SR Path itself may correspond to an ECMP or TE Path.¶

6.1. Local optimization

Local optimization refers to updates applied to a single Junction node that can be performed without requiring coordinated updates across multiple nodes in the DAG. These operations are considered safe when they do not introduce forwarding loops or cause reachability interruptions within the DAG.¶

Examples of local optimizations include:¶

• Manipulating the weight distribution of the outgoing SID lists¶

• Replacing an existing SID list with a more optimal one (e.g. using a new adjacency SID or updated SR path to the next Junction node).¶

• Adding a new SID list toward an egress node that is not itself a Junction node.¶

• Removing a SID list, provided that at least one other outgoing SID list remains active.¶

• Adding a new SID list to a Junction node that connects to a new downstream sub-graph which is not yet connected to the DAG structure.¶

While these operations are scoped to a single node, their correct application may still require awareness of the surrounding DAG structure to avoid introducing loops or reachability issues. For instance, the addition of a new SID list may require confirmation that the new sub-graph is not already part of the DAG, or that it connects loop free.¶

Local optimizations may also be chained in sequence across multiple nodes. When each step maintains loop free and reachable properties, such a chain of updates is still considered local in nature.¶

For example:¶

• Cleaning up an upstream disconnected sub-graph following the removal of an upstream SID list (BSID is no longer used)¶

It is important to note that local optimization differs from global optimization in that it does not require versioning or re-signaling of the entire DAG structure.¶

Since a Junction Segment is realized via an SR Policy, the controller leverages existing protocol mechanisms to update a Junction segment forwarding instructions, as the Junction Segment itself is represented as a candidate path within the SR Policy. When updating an existing candidate path, the binding SID MUST NOT change, as doing so would cause traffic using that binding SID to drop until upstream consumers are updated. Such a change should instead be treated as a global optimization, not a local one.¶

Multiple candidate paths could be used to support more comprehensive local optimizations. However, caution is required in how the MPTED version is encoded and how version increments are signaled, since candidate paths are signaled within the context of the SR Policy.¶

If the same color value is reused and a new candidate path is deployed, the new candidate path will not be installed in the forwarding plane until the existing one is removed, as only one candidate path may be active per SR Policy. This operation may not be hitless, depending on hardware implementation.¶

Alternatively, if a different color value is used, the new candidate path may be allowed to go active. However, it will still fail to do so since it should be using the same binding SID as the existing candidate path. This results in a collision, rendering the new path ineligible, and may also violate protocol constraints that prohibit such configurations.¶

Therefore, it is RECOMMENDED that local optimization be performed by updating the SID list of the existing candidate path, rather than introducing new candidate paths.¶

Appendix B provides examples of local optimization.¶

6.2. Global optimization

Global optimization of a DAG requires coordinated updates across all participating Junction Nodes. This process follows a make-before-break model, where a new version of the DAG is deployed alongside the existing one. The ingress SR Policy (or policies) is the last element to be updated.¶

Since the Color field of an SR Policy indicates DAG membership, and is managed by the controller, global optimization with make-before-break considerations necessitates deploying new Junction segments. This, in turn, requires the instantiation of new SR Policies with unique Color values. These new SR Policies are deployed to all Junction Nodes participating in the updated DAG instance and MUST be associated with distinct Binding SID values from those of the existing instance.¶

To minimize version churn and both color and label consumption, it is RECOMMENDED that controllers limit the number of concurrently deployed DAG instances for the same tunnel. Under normal conditions, only a single active DAG version should exist. During transitions, at most two versions (the current and the new one) should be present. Color and label values may be reused once globally released.¶

The Binding SID associated with a DAG instance MUST remain constant during local optimizations, but MUST change during global optimizations to avoid the risk of routing loops. Therefore, in the example above, if a Junction Node is participating in DAG #1, version #1 continues to use one binding SID value and version #2 is using a distinct binding SID value on each Junction node.¶

Once all Junction segments are deployed to all Junction nodes, the ingress node is updated with new SID lists referencing the Binding SIDs of the new Junction segments downstream.¶

The Color field on ingress nodes is critical for traffic steering and therefore MUST remain constant during global optimizations. In contrast, Color values used on Junction Nodes MAY be reused or encoded to reflect DAG membership or instance mappings. Similarly, the binding SID on the ingress SR Policy also remains constant.¶

The controller tracks the Color values and their corresponding usage for DAG ID and version. For example:¶

• Color value 500 – DAG #1 – Version #1¶

• Color value 501 – DAG #2 – Version #1¶

• Color value 502 – DAG #1 – Version #2¶

When performing global changes, controllers SHOULD ensure that all Junction Nodes are modified in a coordinated and consistent manner before activating the new DAG on the ingress. If the update process fails partially, the controller SHOULD roll back the new deployment and retain the existing stable DAG version and its Junction Segments to avoid inconsistencies in forwarding behavior.¶

Appendix C provides an example of global optimization.¶

7.1. Control of Function through Configuration and Policy

This document describes using the existing SR Policy construct with a color value representing the DAG MID and version. Since SR Policy color value is originally intended for ingress traffic steering on matched routers, a deployment MUST allocate a color range which will be used for MPTE tunnels.¶

7.2. Information and Data Models, e.g., MIB modules

TODO¶

7.3. Liveness Detection and Monitoring

Liveness detection for an MPTE SR deployment has two scopes: the underlying IP links and the DAG tunnel itself.¶

7.4. Verifying Correct Operation

This section discusses OAM for SR-MPLS ([RFC8287]). Applicability to SRv6 ([RFC9259]) is to be covered in a future revision.¶

As with S-BFD in Section 10.3.2, a single OAM probe from the ingress follows only one hashed path through the DAG and does not exercise every egress SID list on every Junction Node.¶

A controller can collect ping and traceroute results on a per-Junction, per-SID-list basis and aggregate them into a holistic view of the DAG. Each probe targets the path encoded in that SID list, terminating at the next downstream Junction Node or DAG egress node, so that every forwarding segment is verified independently of upstream hashing. The controller may request results on-demand or receive them periodically. The frequency in which to run OAM is deployment specific.¶

Because these tests run independently and not necessarily at the same instant, the aggregated view may contain timing skew between individual measurements. Multiple samples per test are RECOMMENDED.¶

7.5. Requirements on Other Protocols and Functional Components

There are currently no new requirements on other protocols or functional components.¶

7.6. Impact on Network Operation

TODO¶

10.1. Normative References

[I-D.draft-ietf-idr-bgp-ls-sr-policy]

Previdi, S., Talaulikar, K., Dong, J., Gredler, H., and J. Tantsura, "Advertisement of Segment Routing Policies using BGP Link-State", Work in Progress, Internet-Draft, draft-ietf-idr-bgp-ls-sr-policy-17, 6 March 2025, <https://datatracker.ietf.org/doc/html/draft-ietf-idr-bgp-ls-sr-policy-17>.

[I-D.draft-ietf-idr-segment-routing-te-policy]

Previdi, S., Filsfils, C., Talaulikar, K., Mattes, P., and D. Jain, "Advertising Segment Routing Policies in BGP", Work in Progress, Internet-Draft, draft-ietf-idr-segment-routing-te-policy-26, 23 October 2023, <https://datatracker.ietf.org/doc/html/draft-ietf-idr-segment-routing-te-policy-26>.

[I-D.draft-kompella-teas-mpte]

Kompella, K., Jalil, L., Khaddam, M., and A. Smith, "Multipath Traffic Engineering", Work in Progress, Internet-Draft, draft-kompella-teas-mpte-02, 2 March 2026, <https://datatracker.ietf.org/doc/html/draft-kompella-teas-mpte-02>.

[RFC2119]

Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, <https://www.rfc-editor.org/rfc/rfc2119>.

[RFC5440]

Vasseur, JP., Ed. and JL. Le Roux, Ed., "Path Computation Element (PCE) Communication Protocol (PCEP)", RFC 5440, DOI 10.17487/RFC5440, March 2009, <https://www.rfc-editor.org/rfc/rfc5440>.

[RFC7880]

Pignataro, C., Ward, D., Akiya, N., Bhatia, M., and S. Pallagatti, "Seamless Bidirectional Forwarding Detection (S-BFD)", RFC 7880, DOI 10.17487/RFC7880, July 2016, <https://www.rfc-editor.org/rfc/rfc7880>.

[RFC8174]

Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, <https://www.rfc-editor.org/rfc/rfc8174>.

[RFC8231]

Crabbe, E., Minei, I., Medved, J., and R. Varga, "Path Computation Element Communication Protocol (PCEP) Extensions for Stateful PCE", RFC 8231, DOI 10.17487/RFC8231, September 2017, <https://www.rfc-editor.org/rfc/rfc8231>.

[RFC8287]

Kumar, N., Ed., Pignataro, C., Ed., Swallow, G., Akiya, N., Kini, S., and M. Chen, "Label Switched Path (LSP) Ping/Traceroute for Segment Routing (SR) IGP-Prefix and IGP-Adjacency Segment Identifiers (SIDs) with MPLS Data Planes", RFC 8287, DOI 10.17487/RFC8287, December 2017, <https://www.rfc-editor.org/rfc/rfc8287>.

[RFC8402]

Filsfils, C., Ed., Previdi, S., Ed., Ginsberg, L., Decraene, B., Litkowski, S., and R. Shakir, "Segment Routing Architecture", RFC 8402, DOI 10.17487/RFC8402, July 2018, <https://www.rfc-editor.org/rfc/rfc8402>.

[RFC9256]

Filsfils, C., Talaulikar, K., Ed., Voyer, D., Bogdanov, A., and P. Mattes, "Segment Routing Policy Architecture", RFC 9256, DOI 10.17487/RFC9256, July 2022, <https://www.rfc-editor.org/rfc/rfc9256>.

[RFC9259]

Ali, Z., Filsfils, C., Matsushima, S., Voyer, D., and M. Chen, "Operations, Administration, and Maintenance (OAM) in Segment Routing over IPv6 (SRv6)", RFC 9259, DOI 10.17487/RFC9259, June 2022, <https://www.rfc-editor.org/rfc/rfc9259>.

[RFC9552]

Talaulikar, K., Ed., "Distribution of Link-State and Traffic Engineering Information Using BGP", RFC 9552, DOI 10.17487/RFC9552, December 2023, <https://www.rfc-editor.org/rfc/rfc9552>.

[RFC9855]

Bashandy, A., Litkowski, S., Filsfils, C., Francois, P., Decraene, B., and D. Voyer, "Topology Independent Fast Reroute Using Segment Routing", RFC 9855, DOI 10.17487/RFC9855, October 2025, <https://www.rfc-editor.org/rfc/rfc9855>.

10.2. Informative References

[RFC9524]

Voyer, D., Ed., Filsfils, C., Parekh, R., Bidgoli, H., and Z. Zhang, "Segment Routing Replication for Multipoint Service Delivery", RFC 9524, DOI 10.17487/RFC9524, February 2024, <https://www.rfc-editor.org/rfc/rfc9524>.