| Internet-Draft | BGP SR Policy for Path Rate | October 2026 |
| Xiong & Li | Expires 3 May 2027 | [Page] |
In AI/ML compute fabrics and data center interconnects, collective communication patterns (e.g., All-Reduce, All-to-All) generate microbursts that can saturate queuing buffers before reactive congestion control mechanisms take effect. This document specifies a BGP SR Policy extension to carry Path Rate information (e.g., Path Committed Rate and Path Maximum Rate), enabling a central controller to compute and distribute per-segment-list rate limits along with the SR Policy Candidate Path. The headend enforces these rate limits on traffic steered to the corresponding segment list, thereby proactively preventing the microburst-induced congestion before traffic enters the network.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 3 May 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
Segment Routing (SR) [RFC8402] is a source routing paradigm that explicitly indicates the forwarding path for packets at the ingress node. The ingress node steers packets into a specific path according to the Segment Routing Policy (SR Policy) as defined in [RFC9256]. In order to distribute SR policies to the headend, [RFC9830] specifies a BGP mechanism to pass SR Policies and Candidate Paths (CPs) in BGP UPDATE messages.¶
In AI/ML compute fabrics and data center interconnects, collective communication patterns such as All-Reduce and All-to-All generate microbursts that can saturate queuing buffers before reactive congestion-control mechanisms take effect. This causes incast congestion, packet drops, and GPU starvation. Conventional congestion control mechanisms (e.g., PFC, ECN/DCQCN, HPCC) rely on the feedback loops and therefore typically cannot adjust the sending rate in a timely manner to prevent the burst from impacting the network.¶
In such deployments, it is useful to associate a rate limit with each segment list of an SR Policy Candidate Path, so the headend can enforce it on the steered traffic. When a controller computes a rate limit for a segment list, it advertises or updates the corresponding SR Policy CP via BGP. The receiving headend applies the rate limit to traffic steered to that segment list. This document defines the control-plane mechanism to distribute path rate information to the headend via BGP SR Policy.¶
This document uses the following abbreviations:¶
CIR: Committed Information Rate¶
PIR: Peak Information Rate¶
SID: Segment Identifier¶
SRv6: Segment Routing over IPv6¶
This document defines the following terminologies:¶
Path Committed Rate : the target sustainable rate (CIR-like) that the network gurrantees on a path. It is expressed in bytes per second.¶
Path Maximum Rate: the peak rate (PIR-like) that a path can sustain, computed by the controller based on the network information collected via Telemetry. It is expressed in bytes per second. The Path Maximum Rate MUST be greater than or equal to Path Committed Rate.¶
Path Rate: the transmission rate that the headend enforced for traffic steered to a path. It is operated within the range from the Path Committed Rate to the Path Maximum Rate.¶
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.¶
As defined in [RFC9830], an SR Policy consists of one or more Candidate Paths (CPs), each comprising one or more segment lists. A headend receives SR Policy CPs via BGP and selects the active CP according to [RFC9256].¶
+------------------+
+--------|Network Controller| Rate Computation
| +---------+--------+
| ^
| |
Rate enforcement Network Information Collection
| |
v |
+--+--+ +-----------+-----------+ +-----+
|Head |---| Segment Routing |---|End |
|end | | Network Domain | |Point|
+-----+ +-----------------------+ +-----+
Figure 1. SR Policy for Path Rate
¶
The SR Policy for Path Rate is as shown in Figure 1. Each segment list of an SR Policy CP has an associated Path Rate, which is used for rate enforcement that the headend applies to traffic steered to that segment list. The Path Rate is expressed as a tuple (Path Committed Rate, Path Maximum Rate).¶
The controller computes the Path Maximum Rate from the collected network information (e.g., buffer occupancy, RTT) via Telemetry according to the computation procedure specified in [I-D.xiong-pce-pcep-pathrate].¶
As defined in [RFC9830], the SR policy encoding structure is as follows:¶
SR Policy SAFI NLRI: <Distinguisher, Policy-Color, Endpoint>
Attributes:
Tunnel Encaps Attribute (23)
Tunnel Type: SR Policy
Binding SID
Preference
Priority
Policy Name
Explicit NULL Label Policy (ENLP)
Segment List
Weight
Segment
Segment
...
...
¶
As introduced in Section 1, each segment list has an associated Path Rate. The SR Policy encoding with Path Rate information is expressed as below::¶
SR Policy SAFI NLRI: <Distinguisher, Policy-Color, Endpoint>
Attributes:
Tunnel Encaps Attribute (23)
Tunnel Type: SR Policy
Binding SID
Preference
Priority
Policy Name
Explicit NULL Label Policy (ENLP)
Segment List
Weight
Path Rate
Segment
Segment
...
...
¶
The Path Rate sub-TLV is an optional sub-TLV of the Segment List sub- TLV. It MUST NOT appear more than once within a single Segment List sub-TLV. If multiple Path Rate sub-TLVs are present, only the first occurrence is processed; subsequent occurrences MUST be ignored and MUST NOT be considered malformed.¶
The Path Rate sub-TLV has the following format:¶
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 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Path Committed Rate |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Path Maximum Rate |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 2. Path Rate sub-TLV
¶
Where:¶
Type: TBD1 (to be assigned by IANA from the "SR Policy Segment List Sub-TLVs" registry [RFC9830]).¶
Length: 8.¶
Path Committed Rate: A 4-octet field specifying the sustainable rate that the network guarantees for the path. The value is encoded as an IEEE 754 single-precision floating-point value, in units of bytes per second.¶
Path Maximum Rate: A 4-octet field specifying the peak rate that the path can sustain. The value is encoded as an IEEE 754 single-precision floating-point value, in units of bytes per second. Path Maximum Rate MUST be greater than or equal to Committed Rate; otherwise, the sub-TLV is malformed (see Section 6).¶
This document does not introduce new BGP processing procedures beyond those defined in [RFC9830]. The existing procedures for advertisement, reception, and error handling of BGP SR Policy NLRIs in [RFC9830] apply directly to this document. The error handling rules specified in Section 5 of [RFC9830] and [RFC7606] govern the processing of malformed Path Rate sub-TLVs.¶
Typically, but not limited to, the SR Policies carrying Path Rate information are configured by a central controller. After configuration, the SR Policies carrying Path Rate information are advertised by BGP UPDATE messages using the same procedures as defined in [RFC9830].¶
Upon receiving a valid Path Rate sub-TLV within a Segment List sub-TLV, the SRPM (SR Policy Module) as per [RFC9830] at the headend programs a per-segment-list rate limiter (shaper or policer) with the specified Committed Rate and Maximum Rate. The headend SHOULD enforce the following semantics:¶
Sustained traffic rate SHOULD NOT exceed the Path Maximum Rate.¶
The headend SHOULD aim to meet the Path Committed Rate as a long-term average for traffic steered to the segment list.¶
Short-term bursts up to Path Maximum Rate are permitted; the burst size is governed by the headend's shaper implementation and local policy.¶
If a Path Rate sub-TLV is absent from a Segment List sub-TLV, the headend MUST NOT apply any rate limiting specific to this document for that segment list and SHOULD revert to the default behavior as defined by [RFC9830].¶
Whenever the Path Committed Rate or the Path Maximum Rate of a segment list is changed, the controller MUST advertise a new BGP UPDATE message containing the updated Path Rate sub-TLV. The headend processes the update according to the rules in [RFC9830].¶
When a new SR Policy CP with an updated Path Rate becomes the active CP, the SRPM SHOULD perform an update of the per-segment-list rate limiter. The update SHOULD NOT cause packet loss, forwarding pause, or throughput degradation to traffic already steered to that segment list.¶
This document extends the BGP SR Policy encoding defined in [RFC9830] with a Path Rate sub-TLV. The security considerations described in [RFC4271] and [RFC9830] are applicable to this document.¶
This document defines a new Sub-TLV in registries "SR Policy Segment List Sub-TLVs" [RFC9830]:¶
Value Description Reference
---------------------------------------------------------------------
TBD1 Path Rate sub-TLV This document
¶