BGP SR Policy Extensions for Performance-Aware Path Selection

Abstract

To enable the headend node to do performance-aware path selection, this document proposes an extension to the BGP SR Policy protocol by defining a new optional Metric Sub-TLV within the BGP Tunnel Encapsulation Attribute [RFC9012]. The introduced Metric Sub-TLV encodes performance parameters (such as latency, bandwidth, reliability, etc.) for SR Policy paths.¶

This specification also updates the BGP route selection procedures in [RFC4271], modifying the Breaking Ties (Phase 2) logic to prioritize the metrics for SR Policy paths.¶

Key contributions include:¶

• Introduce Metric Sub-TLV in BGP SR Policy¶
• Update the tie-breaking procedure for BGP route selection¶

1. Introduction

Segment Routing (SR) [RFC8402] allows a headend node to steer a packet flow along a specific path. [RFC9256] further details the concepts of SR Policy and steering into an SR Policy. [I-D.ietf-idr-sr-policy-safi] specifies the use of BGP to distribute one or more of the candidate paths of an SR Policy to the headend node of that policy. Currently [I-D.ietf-idr-sr-policy-safi] lacks the capability to propagate performance metrics such as path latency, bandwidth, or reliability. This limitation prevents headend nodes from implementing policy selection based on path metrics when there are multiple paths reaching the same destination. Consequently, the headend nodes cannot dynamically elect performance-optimal path among multiple SR Policies.¶

[I-D.ietf-idr-sr-policy-metric] is intended to address this problem. However, it carries metrics within segment lists, and a single segment list may contain more than one metric. When the candidate path of a SR Policy comprises multiple segment lists, the headend node faces difficulties in determining the metric for that SR Policy. This situation becomes even more complex when each segment list itself includes multiple metrics.¶

Furthermore, merely extending the SR Policy protocol with a metric attribute cannot resolve the current issue. This is because, before the BGP process on the headend node considers multiple alternative SR Policies, it has already determined the next hop of the BGP route in accordance with the existing Tie-breaking Procedure. Subsequently, only the SR Policies leading to the selected next hop may be utilized. Even if there exist more optimal SR Policies that can reach the destination via other next hops, these SR Policies will not be employed. Therefore, based on the content currently provided in [I-D.ietf-idr-sr-policy-metric], it is still impossible to ensure that the optimal SR Policy is selected.¶

To address these limitations, this document extends the BGP SR Policy protocol to carry performance metrics in the candidate path of a SR Policy and updates the BGP Tie-breaking Procedure to prioritize metrics for SR Policy paths.¶

2. Use Case

As illustrated in Figure 1, the SR Policy Computation and Provisioning System, such as a SDN controller, collects real-time network state information (e.g., topology, link bandwidth) and performance metrics (e.g., link latency, jitter, packet loss rate) via BGP-LS [RFC7752], TWAMP [RFC5357]/Micro TWAMP [RFC9533], and Telemetry [RFC9232] etc.. Based on service or customer requirements (e.g., minimum latency), the system computes SR Policy paths between designated endpoints and distributes these paths to the headend nodes via the BGP SR Policy protocol [I-D.ietf-idr-sr-policy-safi].¶

For example:¶

The system provisions two low-latency policies to headend node PE3:¶

Policy1: Path via P1-->PE1, with a measured latency of 20 ms.¶

Policy2: Path via P2-->PE2, with a measured latency of 12 ms.¶

However, the current BGP SR Policy protocol [I-D.ietf-idr-sr-policy-safi] only propagates path definitions (e.g., segment lists) without embedding performance metrics. This forces headend nodes to select paths based solely on static criteria (e.g., administrative preferences), potentially leading to suboptimal traffic engineering decisions.¶

To address this limitation, this proposal extends the BGP SR Policy protocol by introducing a new Performance Metrics Sub-TLV within the BGP Tunnel Encapsulation Attribute [RFC9012]. This Sub-TLV encodes key performance indicators (KPIs) such as latency, bandwidth, and reliability (see Section 3 for details). With this extension: The SR Policy Computation and Provisioning System can advertise SR Policies alongside their associated KPIs. Headend nodes leverage the enhanced BGP route selection logic (Section 6) to prioritize paths that meet dynamic performance requirements.¶

       +---------------------------------------------+
       |SR Policy Computation and Provisioning System|
       +---------------------------------------------+
                *       *            *        *
               *       *              *        *Extended BGP
              *       *                *        *SR Policy
             *       *                  *        *
         +---+  15ms*   +---+       5ms  *      +---+
        /|PE1| ----*----| P1|-------------*-----|PE3|\
       / +---+    *     +---+----\      /--*----+---+ \
      /   |      *        |       \    /    *     |    \
+---+/    |     *         |        ----\     *    |     \+---+
|CE1|0.5ms|    *          |0.5ms  /     \     *   |0.5ms |CE2|
+---+\    |   *           |   ----       ---\  *  |     /+---+
      \   |  *            |  /1ms         5ms\  * |    /
       \ +---+          +---+                 --+---+ /
        \|PE2|----------|P2 |-------------------|PE4|/
         +---+   11ms   +---+       1ms         +---+

3. Extensions to BGP SR Policy

This document extends the BGP SR Policy protocol by introducing a new sub-TLV, Metric Sub-TLV, within the BGP Tunnel Encapsulation Attribute. The Extended BGP SR Policy Encoding structure is as follows:¶

        SR Policy SAFI NLRl: <Distinguisher, Policy-Color, Endpoint>
        Attributes:
          Tunnel Encapsulation Attribute(23)
            Tunnel Type: SR Policy(15)
              Binding SID
              SRv6 Binding SID
              Preference
              Priority
              Metric
              Policy Name
              Policy Candidate Path Name
              Explicit NULL Label Policy (ENLP)
              Segment List
                Weight
                Segment
                Segment
                ...
              ...

Metric Sub-TLV is used to carry performance metrics such as latency, bandwidth, and reliability. The format of the Metric Sub-TLV is as follows:¶

 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|      Type     |    Length     |     Flags     |    Reserved   |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                 Delay(8 octets,optinal)                       |
|                                                               |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|              Bandwidth(4 octets,optional)                     |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|             Reliability(4 octets,optional)                    |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Where:¶

Type (1 octet): Indicates this sub-TLV is Metric, Specific values need to be assigned by IANA.¶

Length (1 octet): Indicates the length of the Metric Sub-TLV in bytes.¶

Flags (1 octet): Indicates the presence of specific performance metrics. Its definition is shown in Figure 4.¶

Reserved (1 octet): Reserved for future use. This field MUST be set to 0 when sending and ignored when receiving.¶

Delay(8 octets): Carries delay information. Its format depends on the D flag in the Flags Field:¶

• If D = 01: NTPv4 format delay¶
• If D = 10: IEEE 1588v2 PTP format delay¶

Bandwidth (4 octets): Carries bandwidth information in Mbps.¶

Reliability (4 octets): Carries reliability information, such as the maximum number of failures that have occurred on all links in the SR Policy path within the past month.¶

 0 1 2 3 4 5 6 7
+-+-+-+-+-+-+-+-+
| D |B|R|Reserved|
+-+-+-+-+-+-+-+-+

Where, D Flag is for delay, B Flag is for Bandwidth and R Flag is for reliability, all the other bits are reserved. The detailed encodings of the three flags defined in this document are as follows: ‌¶

+------+-------+-------------------------------+
| Flag | Bits  | Description                   |
+------+-------+-------------------------------+
| D    | 0-1   | 00: No delay                  |
|      |       | 01: NTPv4 delay               |
|      |       | 10: PTP delay                 |
|      |       | 11: Reserved                  |
+------+-------+-------------------------------+
| B    | 2     | 0: No bandwidth               |
|      |       | 1: Bandwidth                  |
+------+-------+-------------------------------+
| R    | 3     | 0: No reliability             |
|      |       | 1: Reliability                |
+------+-------+-------------------------------+

Implementations SHOULD set only one flag (D, B, or R) at a time, as these metrics are typically not directly comparable. Network operators MAY configure which metric to prioritize based on service requirements.¶

4. Policy Computation and Provisioning System Behavior

The Policy Computation and Provisioning System is responsible for calculating Segment Routing (SR) policies based on network state and business requirements, and provisioning them to headend nodes. When provisioning SR policies that include performance metrics, the system should follow these steps:¶

Collect Network State Information: Gather real-time network topology, link utilization, and other relevant data.¶

Compute SR Policies: Calculate SR Policy paths that meet performance requirements based on service needs and network state.¶

Encapsulate Performance Metrics: Embed performance metrics such as latency, bandwidth, and reliability within the Metric Sub-TLV of the BGP Tunnel Encapsulation Attribute.¶

Provision BGP Update Messages: Include the SR policies with performance metrics in BGP update messages and send them to the appropriate headend nodes.¶

5. Headend Node Behavior

Upon receiving SR policies with performance metrics, headend nodes should process them as follows:¶

Parse BGP Update Messages: Extract SR policies and their associated performance metrics from the received BGP update messages.¶

Store Performance Metrics: Save the performance metrics in a local database for subsequent path selection.¶

Path Selection: Prioritize paths that meet dynamic performance requirements when multiple paths are available.¶

Update Routing Tables: Modify routing tables based on the selected paths to ensure traffic is forwarded along optimized routes.¶

6. Updated Tie-Breaking Procedure for BGP

Support for SR Policy metric introduced in this document involves several modifications to the tie-breaking procedures of the BGP "phase 2" decision described in [RFC4271], Section 9.1.2.2.¶

A new step, step e0, is inserted before step e of the tie-breaking (Phase 2) logic in [RFC4271].¶

d) If at least one of the candidate routes was received via EBGP, remove from consideration all routes that were received via IBGP.¶

e0) If any routes have the Color extended community attribute with identical values, remove from consideration all routes lacking the Color extended community attribute.¶

Compare the SR policies corresponding to the remaining routes. If any SR policies have the Metric Sub-TLV, remove from consideration all routes whose corresponding SR policies lack the Metric Sub-TLV.¶

Remove from consideration all routes whose Metric Sub-TLV are not the best. For the latency metric, the smallest value is considered the best; for the bandwidth metric, the largest value is the best; and for reliability, the smallest value is the best.¶

e) Remove from consideration any routes with less-preferred interior cost.¶

9. References