Internet-Draft MPLS OPT MNA Flag for OAM November 2026
Song, et al. Expires 22 May 2027 [Page]
Workgroup:
MPLS
Internet-Draft:
draft-ietf-mpls-on-path-telemetry-flag-06
Published:
Intended Status:
Standards Track
Expires:
Authors:
H. Song
Futurewei Technologies
G. Fioccola
Huawei Technologies
R. Gandhi
Cisco Systems

MPLS On-Path Telemetry Network Action Flag for OAM

Abstract

This document describes postcard-based on-path telemetry with packet marking (PBT-M) using an MPLS Network Actions (MNA) flag to support Operations, Administration, and Maintenance (OAM) in MPLS networks. The scheme uses a single flag bit carried in a Flag-Based Network Action Indicator (Opcode 1) of the MNA Sub-Stack as defined in RFC 9994. In addition to addressing the protocol requirements for applying PBT-M, this document provides comprehensive operational, manageability, and security considerations.

Status of This Memo

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 22 May 2027.

▲

Table of Contents

1. Introduction

To gain detailed data plane visibility to support effective network OAM, it is essential to be able to examine the trace of user packets along their forwarding paths. Such on-path flow data reflect the state and status of each user packet's real-time experience and provide valuable information for network monitoring, measurement, and diagnosis.

The telemetry data include, but are not limited to, the detailed forwarding path, the timestamp/latency at each network node, and, in case of packet drop, the drop location as well as the reason. The emerging programmable data plane devices allow user-defined data collection or conditional data collection based on trigger events. Such on-path flow data are from and about the live user traffic, which complements the data acquired through other passive and active OAM mechanisms such as IPFIX [RFC7011] and ICMP [RFC0792].

On-path telemetry was developed to cater to the need of collecting on-path flow data. There are two basic modes for on-path telemetry: the passport mode (e.g., IOAM trace option [RFC9197]) and the postcard mode (e.g., IOAM direct export option (DEX) [RFC9326]).

In MPLS networks, MPLS Network Action (MNA) [RFC9789] extends the MPLS label stack by supporting extra in-stack network actions and ancillary data encoded in stack. The in-stack MNA Sub-Stack is described in [RFC9994]. MNA also extends the MPLS payload by supporting extra post-stack network actions and ancillary data encoded post-stack. The post-stack MNA header is described in [I-D.ietf-mpls-mna-ps-hdr].

This document describes the method to apply a new variation of the postcard mode on-path telemetry, PBT-M, to MPLS networks using an MNA flag only. PBT-M does not require a telemetry instruction header but a single trigger bit in MNA flags. A similar mechanism has been adopted for SRv6 OAM [RFC9259], which uses the O-bit in SRH flags as the marking bit to trigger on-path telemetry. The key benefits of PBT-M are its low overhead and high flexibility. However, extracting telemetry data in this manner introduces unique protocol requirements, alongside critical operational and manageability challenges that must be addressed for real-world deployment. This document discusses these requirements and provides comprehensive solutions for MPLS networks. Crucially, it outlines the operational guidelines necessary for safe deployment, including concrete mechanisms for load control, DoS mitigation, configuration scalability, and managing telemetry across partially upgraded forwarding paths.

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.

1.2. Terminology

This document uses the following terms:

Head-end node:
The node at the start of the monitored LSP or SR path that imposes the MNA Sub-Stack and sets the P-flag on selected packets. It is the only node that marks packets.
Transit node:
A node on the monitored path between the head-end node and the egress node.
Egress node:
The node at the end of the monitored path. It is within the Hop-by-Hop scope of the PBT-M action and removes the PBT-M network action if necessary.
PBT-M-aware node:
A head-end, transit, or egress node that supports PBT-M.

2. PBT-M: Direct Export for On-path Telemetry based on Packet Marking

As the name suggests, PBT-M only needs a marking-bit in the existing headers of user packets to trigger the telemetry data collection and export. The sketch of PBT-M is as follows. If on-path data need to be collected, the user packet is marked at the head-end node. At each PBT-M-aware node, if the mark is detected and data collection is enabled, a postcard packet (i.e., the dedicated OAM packet triggered by a marked user packet) is generated and sent to a collector. The postcard contains the data configured by the management plane. Once the collector receives all the postcards for a single user packet, it can infer the packet's forwarding path and analyze the data set. The egress node is configured to remove the marking from the packets and restore their original format if necessary.

The overall architecture of PBT-M is depicted in Figure 1.

                      +------------+        +-----------+
                      | Network    |        | Telemetry |
                      | Management |(-------| Data      |
                      |            |        | Collector |
                      +-----:------+        +-----------+
                            :                     ^
                            :configurations       |postcards
                            :                     |(OAM pkts)
             ...............:.....................|........
             :             :               :      |       :
             :   +---------:---+-----------:---+--+-------:---+
             :   |         :   |           :   |          :   |
             V   |         V   |           V   |          V   |
          +------+-+     +-----+--+     +------+-+     +------+-+
usr pkts  |Head-end|     |Transit |     |Transit |     | Egress |
     ====>| Node   |====>| Node   |====>| Node   |====>| Node   |===>
          |        |     | A      |     | B      |     |        |
          +--------+     +--------+     +--------+     +--------+
        mark usr pkts  gen postcards  gen postcards  gen postcards
        gen postcards                                unmark usr pkts
Figure 1: Architecture of PBT-M

The advantages of PBT-M are summarized as follows.

3. New Requirements

Although PBT-M has some unique features, it also introduces a few new requirements.

4. Design Considerations

To address the above requirements, this section specifies several design details for applying PBT-M in MPLS networks.

4.1. Packet Marking

To trigger the path-associated data collection, usually, a single bit from some header field is sufficient. The proposed action encoding is shown in Figure 2 using the MNA Sub-Stack formats defined in [RFC9994].

    0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |         MNA-Label = bSPL (4)          | TC  |S|      TTL      | (A)
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |  Opcode=1   |         Data=0          |R|IHS|S|NASL=1 |U|NAL=1| (B)
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |1|                  Data=0                   |S|P|      0      | (D)
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 2: Action Encoding

In the figure, three LSE formats defined in [RFC9994] are used to form the Network Action Sub-Stack (NAS). Format A (labeled A) is the MNA Sub-Stack Indicator carrying the MNA bSPL (value 4). Format B (labeled B) is the Initial Opcode LSE; it uses the Flag-Based Network Action Indicators without Ancillary Data opcode (Opcode 1, Section 6.2 of [RFC9994]), sets the scope (IHS) for the whole NAS, and points (NAL=1) to one following Format D LSE. Format D (labeled D) carries the PBT-M indicator (P-flag). No Format C LSE is needed, since a Format D LSE only requires a preceding Format A and Format B LSE (Section 4.4 of [RFC9994]).

The PBT-M indicator (P-flag) is carried as a single flag in the Format D data field. If the bit is set to '1', every PBT-M-aware node that receives the packet is triggered to collect and export the telemetry data as configured by the management plane. Within the Hop-by-Hop scope of the PBT-M action, this includes the head-end node that originates the flag, every on-path transit node, and the egress node. A node with data collection enabled generates and exports one postcard per marked packet as configured by the management plane.

Following Section 10 of [RFC9994], the PBT-M network action is defined as follows:

Because PBT-M marks only a subset of the packets in a flow, the value of the P-flag differs among packets of the same flow. Per Section 5.2 of [RFC9994], such mutable data MUST NOT be placed in the most significant 20 bits (the label value, which is always used for ECMP), nor in the most significant 23 bits when the TC field is also used for ECMP hashing, of a Format B, C, or D LSE; otherwise it may perturb ECMP load-balancing and cause out-of-order delivery within a flow. The P-flag MUST therefore be assigned a bit position that lies outside both the most significant 20 bits and the most significant 23 bits of the LSE, i.e., within the trailing data bits of the Format D LSE (word bits 24-31), so that the marking bit does not affect ECMP load-balancing whether or not the TC field is used in the hash. The concrete bit position is to be allocated by IANA (see Section 7) and is referred to as TBD1 in this document; the position shown for the P-flag in Figure 2 is illustrative only.

4.2. Flow Path Discovery

In case the path that a flow traverses is unknown in advance, all PBT-M-aware nodes should be configured to react to the marked packets by exporting some basic data, such as node ID and per-LSE TTL (see Section 4.3) before a data set template for that flow is configured. This way, the management plane can learn the flow path dynamically.

If the management plane wants to collect the on-path data for some flow, it configures the head-end node with a marking interval (the default, see Section 4.4) or, optionally, a marking probability. When the first marked packet is forwarded in the network, the PBT-M-aware nodes will export the basic data set to the collector. Hence, the flow path is identified. If other data types need to be collected, the management plane can further configure the data set's template to the target nodes on the flow's path. The PBT-M-aware nodes collect and export data accordingly if the packet is marked and a data set template is present.

To mitigate configuration churn caused by frequent path shifts (e.g., due to dynamic routing or ECMP), the basic export data set SHOULD remain enabled as a standing, flow-independent default on all PBT-M-aware nodes, so path discovery needs no per-flow state and follows a shift automatically. The per-flow detailed templates SHOULD be installed as soft state with an idle timeout that is refreshed by continued marking and expires automatically rather than being explicitly revoked on each path change. The idle timeout SHOULD be set to a small multiple of the expected inter-mark time for the flow (e.g., ten times the marking interval). To absorb ECMP or fast-reroute shifts without a controller round-trip, the controller MAY pre-install these templates on the ECMP-group members and immediate next-hops adjacent to the current path.

4.3. Packet Identity for Export Data Correlation

The collector needs to correlate all the postcard packets for a single user packet. Once this is done, the hop order along the forwarding path can be inferred from the TTL, or from the timestamp if the network time is synchronized. Note, however, that an MPLS packet carries a label stack in which each LSE has its own TTL and only the top LSE TTL is decremented at each hop; PUSH and POP operations change which LSE is on top, so the top-of-stack TTL is not monotonic end-to-end and may, for example, increase after a POP. A single TTL value is therefore insufficient in the general case. A PBT-M-aware node SHOULD instead report the TTL of every LSE in the received label stack (the per-LSE TTL vector) together with its node ID, and the collector reconstructs the hop order from this TTL vector, using IGP topology where needed to resolve label-stack operations along the path. Reporting the received label stack itself is not required for hop ordering; it is useful mainly as the flow ID and for detecting PUSH/POP operations, and the swapped top label is in any case not a stable per-hop identifier. The key issue here is to correlate all the postcards for the same user packet.

The first possible approach includes the flow ID in the OAM packets. In case of MPLS, if we consider the aggregated flow sharing the LSP/FEC, the MPLS label stack can serve as the flow ID; if a 5-tuple IP flow needs to be identified, data beyond the label stack needs to be collected to serve as flow ID. If the packet marking interval is large enough, the flow ID can be used to identify a user packet with high likelihood. As a result, all the exported postcard packets for the same flow during a short time interval likely belong to the same user packet.

Alternatively, if the network is time-synchronized, then the flow ID plus the timestamp at each node can also infer the postcard affiliation. This requires the participating nodes to share a common time reference (e.g., using PTP or NTP) with a synchronization accuracy that is small relative to the minimum inter-hop latency along the path. Timestamps carried in postcards use a single global epoch and a format that is independent of local time zones (e.g., the PTP or NTP timestamp formats defined in [RFC9197]). Even so, some errors may occur under some circumstances. For example, if two consecutive user packets from the same flow are marked and one exported postcard from a node is lost, the collector may be unable to decide to which user packet the remaining postcard is related. Postcards that cannot be unambiguously correlated MUST be discarded rather than reported as part of a path.

Guaranteed correlation, which is the default behavior (Section 4.4), is achieved by ensuring that at most one marked packet is in flight at a time per correlation key that is actually carried in the postcards, i.e., per LSP/label stack when only the label stack is exported, or per finer-grained flow ID when additional identifying data is exported. Because only the head-end node marks packets (see Section 6), the head-end node enforces this by spacing the marks so that the marking interval exceeds the maximum in-network packet lifetime plus the maximum postcard export delay. Under this constraint, the carried flow ID correlates all postcards of a single marked packet unambiguously, even under loss or reordering, at the cost of a lower marking rate. A postcard set with gaps (e.g., due to postcard loss or a node skipping the trigger at its postcard cap) is then still attributable to a single marked packet and SHOULD be reported as an incomplete path rather than discarded. A PBT-M-aware node SHOULD discard a postcard that cannot be exported within the maximum postcard export delay and count the discard.

Correlation requires all postcards of a single user packet to be evaluated together. Therefore, if postcard processing is distributed across multiple collectors for load balancing (Section 4.4), all postcards that share a correlation key MUST be steered to the same collector (for example, by hashing the correlation key), or to a common correlation function that aggregates them; otherwise the postcards of a single user packet cannot be brought together and the path cannot be reconstructed.

4.4. Load Control

PBT-M MUST be applied to only a small subset of packets, never to all packets all the time. It is better used in an interactive environment where the network telemetry applications dynamically decide which subset of traffic is under scrutiny. To bound the DoS vector identified in Req. 4, PBT-M implementations MUST enforce two independent rate limits, and MUST ship with a conservative default posture that is enabled without operator action and that operators MAY tune to their platform and deployment:

  • Marking rate (at the head-end node): the default marking mode is interval-based. The head-end node marks at most one packet per correlation key (Section 4.3) per marking interval, where the interval MUST exceed the maximum in-network packet lifetime plus the maximum postcard export delay. The RECOMMENDED default interval is 1 second. The head-end node's aggregate marking rate across all correlation keys MUST also be bounded and SHOULD be set so that the total postcard rate expected at any on-path node stays below that node's postcard cap, so that postcard drops are the exception rather than the normal state. The resulting overhead is quantified in the overhead model below; operators SHOULD re-evaluate it for their own deployment. Other marking modes (e.g., probabilistic marking) MAY be configured. In this case, the correlation guarantee of Section 4.3 does not hold, and postcards that cannot be unambiguously correlated are discarded.

  • Postcard generation rate (at every PBT-M-aware node): the number of postcards a node generates MUST be capped, e.g., by a token bucket. The RECOMMENDED default is an average of 1000 postcards per second per node with a burst of 2000, above which the trigger is skipped and the drop is counted rather than queued. These values are conservative starting points that have not been validated by deployment experience; operators are expected to calibrate them against node forwarding capacity and collector ingest capacity.

  • Control-plane protection: postcard generation MUST NOT degrade the node's control-plane functions, in particular IGP (e.g., IS-IS) adjacency maintenance and link-state flooding, whose correct and timely operation is more important than telemetry export. Because postcard generation may be performed on a punt or slow path that may share resources with control-plane processing, such an implementation MUST schedule it at a priority strictly below control-plane traffic and MUST subordinate its CPU, queue, and punt-path usage to control-plane needs. When contention occurs, postcard generation MUST be throttled or skipped (and the drop counted, see Section 5.4) in favor of control-plane work. Operators SHOULD set the postcard generation cap so that, even under worst-case marking, measured control-plane performance (e.g., IS-IS flooding rate and convergence time) is not affected.

The postcard overhead can be estimated as follows. Let T be the marking interval, N the number of PBT-M-aware nodes on the path that each generate a postcard for a marked packet, R the flow's packet rate, Sc the mean size of the monitored (customer) packets, Sp the mean size of a generated postcard including its own transport encapsulation and the exported data, and C the per-node postcard cap. The flow's marking rate is r = min(1/T, R), i.e., one mark per interval, or every packet for a flow slower than one packet per interval. A single marked packet triggers up to N postcards. Consequently, each on-path node generates about r postcards per second for the flow; the network-wide postcard rate for the flow is about N*r; the added byte load relative to the flow is about (r/R)*(Sp/Sc) on any single link and about N*(r/R)*(Sp/Sc) network-wide. The total postcard rate at a node is the sum of r over all monitored correlation keys whose marked packets traverse that node, which is at most K/T for K such keys. Keeping it below the cap therefore requires K < C*T. The collector ingest is the sum of N*r over all monitored keys. As a worked example, with T=1s, N=8, R=1,000,000 packets/s, Sc=1,000 octets, and Sp=128 octets, r = 1 mark/s, so each node emits about 1 postcard/s for this flow, the network-wide postcard rate is about 8 packets/s, and the network-wide byte overhead is about 0.0001% of the flow. If a head-end node monitors 200 such flows over the same path, each on-path node emits about 200 postcards/s, i.e., 20% of the default cap of 1,000 postcards/s, and the collector receives about 1,600 postcards/s (about 1.6 Mbit/s). With the default T and C, the cap is reached only when more than 1,000 monitored correlation keys traverse a single node. Because the marks of different keys may coincide, a head-end node SHOULD spread the marks of different keys across the interval (e.g., by applying a random offset to each key's marking schedule) so that the bursts seen by on-path nodes stay within the token-bucket burst size. Because these quantities scale linearly with the number of monitored keys, 1/T, N, and Sp, an operator SHOULD recompute them for its own deployment and set the marking interval, the set of monitored keys, and the per-node cap accordingly.

Marked packets in excess of these limits MUST be forwarded normally but do not trigger a postcard. Because each node applies its cap independently, different nodes may skip different marked packets, leaving gaps in the postcard sets. Under the default marking mode, such a set is still unambiguously correlated (Section 4.3), and the dropped-postcard counters (Section 5.4) allow it to be distinguished from a partial path. The postcard packets can be distributed to different collectors to balance the processing load, subject to the correlation constraint in Section 4.3: all postcards sharing a correlation key MUST be delivered to the same collector or to a common correlation function.

Because PBT-M sends telemetry data by dedicated postcard packets, it allows data aggregation and compression before exporting from the on-path nodes. Each node can process the generated raw data according to the configured local data-export policies. Such policies may specify how raw data is used to calculate performance metrics, e.g., max, min, mean, percentile, etc.

Domain boundary handling (ingress filtering): PBT-M is scoped to a single trust domain (see Section 6). A node at the boundary of that domain MUST, on interfaces that face outside the domain, remove any PBT-M network action (i.e., clear the P-flag) from packets entering the domain, or drop such packets. This ensures the P-flag can be set only by the domain's own head-end node and prevents an external party from injecting marked packets to trigger postcard generation. Within the domain, the two rate limits above apply: the marking-rate limit at the head-end node, and the postcard-generation-rate cap at every PBT-M-aware node.

5. Operational Considerations

5.1. Operational and Manageability Mechanisms

Following the baseline set by [I-D.ietf-opsawg-rfc5706bis], [RFC9994], and [I-D.ietf-mpls-mna-ps-hdr], PBT-M implementations MUST support the following operational mechanisms:

  • Counters: Nodes MUST maintain the counters listed in Section 5.4.

  • Rate-limited Alarms: Nodes SHOULD generate rate-limited alarms when postcard dropping exceeds a configured threshold or when anomalous PBT-M flag rates are detected.

  • Correlation Reliability: The confidence with which postcards are associated to a single user packet depends on the marking discipline and, optionally, on time synchronization, as specified in Section 4.3. For deterministic behavior, operators SHOULD use the single-marked-packet-in-flight discipline of Section 4.3, which is the default marking mode (Section 4.4); postcards that cannot be unambiguously correlated are discarded rather than reported as a path.

  • Partial Path Support: In deployments where not all nodes are PBT-M-aware, the collector will receive an incomplete postcard set. Consistent with the capable/incapable node interaction described in Section 12.3 of [RFC9994], the head-end node sets the U-bit to zero, so a non-PBT-M-aware node skips the action and forwards the packet without generating a postcard. To keep the resulting gaps interpretable, operators and analytic systems SHOULD use TTL gaps or IGP topology data to locate missing hops, node capability information to distinguish a PBT-M-aware hop that is idle (aware, but exported no data) from a non-PBT-M-aware hop that cannot respond to the trigger at all, and the dropped-postcard counters to identify hops that skipped the trigger at their postcard cap.

  • Interaction with MSD, RLD, and MTU: PBT-M imposes an MNA Sub-Stack of three LSEs (Figure 2) on marked packets. On platforms where imposing this Sub-Stack reduces the label-stack depth otherwise available to SR-MPLS, a node that enables PBT-M MUST reflect the reduction in the Maximum SID Depth (MSD) it advertises per [RFC8476][RFC8491], decrementing the advertised value by the number of LSEs PBT-M consumes (up to three), so that a controller (e.g., a centralized path computation element) can correctly determine whether a given SID stack is supported on that node. Because three label positions may be significant on some platforms, operators SHOULD account for this budget when planning SR Policies on PBT-M-enabled paths. A node on which imposing the PBT-M Sub-Stack does not reduce the label-stack depth available to SR-MPLS need not change its advertised MSD. In addition, every PBT-M-aware transit node must be able to read the label stack down to the PBT-M MNA Sub-Stack in order to detect the P-flag; the Sub-Stack MUST therefore be placed within the Readable Label Depth (RLD) of the on-path MNA-capable nodes (see [RFC9994]). Operators SHOULD account for the RLD needed to reach the PBT-M Sub-Stack when positioning it in the stack, since a node whose RLD does not reach the Sub-Stack cannot act on the trigger and will silently behave as a non-PBT-M-aware hop. Imposing the PBT-M Sub-Stack also increases the size of a marked packet by up to 12 octets at the head-end node, with no further increase along the path. Operators SHOULD ensure that the path MTU accommodates this increase, and a head-end node SHOULD NOT mark a packet whose size after imposition would exceed the path MTU, so that marking does not cause drops or fragmentation of the monitored traffic.

  • Verifying Correct Operation: Before relying on PBT-M in production, an operator SHOULD confirm correct operation by marking a small, controlled set of test packets on a known path and verifying that the expected postcard set (one per PBT-M-aware node on that path) is received at the collector.

  • Management Data Model: This document does not define a YANG data model for configuring or monitoring PBT-M. Section 5.4 provides a reference information model to guide a future data model.

5.2. Configuration

PBT-M is disabled by default. If enabled, access lists with an optional sampler, [RFC5476], should be configured and attached at the ingress interface of the head-end node to select the intended flows for PBT-M.

Based on the requirements and node capability, the flow data could be exported at the head-end node, each transit node, and the egress node with IPFIX [RFC7011].

5.3. Data Export

The data decomposition can be achieved on the PBT-M-aware node exporting the data or on the IPFIX data collection. [I-D.spiegel-ippm-ioam-rawexport] describes how data is being exported when decomposed at IPFIX data collection. When being decomposed on the PBT-M-aware node the data can be aggregated according to section 5 of [RFC7015].

Postcard packets MUST NOT carry the PBT-M network action, and a PBT-M-aware node MUST NOT generate a postcard in response to a postcard.

Postcards can be transported out-of-band (e.g., over a dedicated management network) or in-band, sharing links and queues with the monitored traffic. For in-band transport, it is RECOMMENDED that the traffic class (MPLS TC or DSCP) used for postcards be configurable. In choosing it, operators trade off two goals: a low-priority class minimizes the impact on user traffic, but postcards are then more likely to be lost on a congested or degraded path; a higher-priority class improves delivery but competes with user traffic. Since the postcard volume is bounded by the rate limits, a class at or above that of the monitored traffic is usually affordable. Where diagnosis of congested paths is required, operators SHOULD use such a class or out-of-band transport.

The postcard packet format and the full set of exportable data are outside the scope of this document. However, for the correlation and path reconstruction procedure to work across implementations, every postcard SHOULD contain at least the following information:

When IPFIX is used as the export encoding, existing Information Elements are sufficient for this minimum set, with each postcard exported as a per-packet record: the label stack, including all TTL fields, can be carried in mplsLabelStackSection (ElementId 316) [RFC5477], with a length covering the label stack down to the bottom-of-stack LSE; the node identifier in exporterIPv4Address or exporterIPv6Address (ElementId 130 or 131), set to an address of the node that is advertised in the IGP; data beyond the label stack in mplsPayloadPacketSection (ElementId 317) [RFC5477]; and the timestamp in observationTimeNanoseconds (ElementId 325) [RFC5477], which uses the NTP timestamp format [RFC7011].

Because postcard export protocols can be extensible and telemetry ingestion operates at high scale, collectors MUST apply defensive input validation to postcard headers and payloads, as detailed in Section 6.

5.4. Reference Information Model

This section summarizes the PBT-M configuration parameters and counters specified in this document, as a basis for a future data model.

Table 1: PBT-M Configuration Parameters
Parameter Scope Default Ref.
PBT-M enable node, interface disabled 5.2
Flow selection (access list, optional sampler) head-end none 5.2
Marking mode head-end interval-based 4.4
Marking interval T head-end 1 s 4.4
Aggregate marking rate head-end deployment-specific (K < C*T) 4.4
Basic data set (node ID, label stack with TTLs) node enabled with PBT-M 4.2, 5.3
Per-flow data set templates node none 4.2
Template idle timeout node small multiple of T 4.2
Postcard cap rate / burst node 1000/s / 2000 4.4
Maximum postcard export delay node implementation-specific 4.3, 4.4
Postcard traffic class node implementation-specific 5.3
Collector address(es) and steering node none 4.3, 4.4
Domain-boundary interfaces interface none 4.4
Boundary filter action (clear P-flag or drop) interface implementation-specific 4.4
Alarm thresholds node implementation-specific 5.1

A PBT-M node maintains the following counters:

  • Marked packets imposed: per head-end marking policy (flow selection).

  • Marked packets received: per interface.

  • Postcards generated: per node and per data set template.

  • Postcards dropped, per node and reported separately for each cause: (a) the postcard cap was exceeded; (b) generation throttled or skipped by control-plane protection; (c) export failed (e.g., the collector is unreachable or the transport session is down) or the postcard could not be exported within the maximum postcard export delay.

  • Malformed MNA instructions: per interface.

  • Boundary-filtered packets: packets whose P-flag was cleared or that were dropped by domain-boundary ingress filtering per boundary interface, reported separately for clear and drop.

6. Security Considerations

In the intended operation, only the head-end node of the monitored LSP/SR path sets the P-flag, and on-path nodes only read it and react by generating a postcard. This is an authorization expectation, not a property enforced by the encoding: because the P-flag is a mutable, unauthenticated bit processed hop-by-hop, any on-path node that is compromised or misconfigured can set, clear, or otherwise alter it. Setting or spoofing the flag can drive excess postcard generation, a DoS vector against both the nodes and the collectors; clearing or flipping it can suppress or falsify telemetry, yielding unreliable measurements.

This document does not provide cryptographic integrity protection for the flag. Protecting a single per-packet-mutable bit, per hop and in-band, is impractical and would defeat the low-overhead goal of PBT-M; this property is shared with other lightweight marking triggers such as the SRv6 O-bit [RFC9259] and IOAM DEX [RFC9326]. PBT-M therefore relies on the trust model of the MPLS/MNA domain, which MUST be enforced at the domain boundary. Specifically: ingress filtering at the domain boundary, as defined in Section 4.4, MUST clear or drop the P-flag on packets entering from outside the domain, so that only the domain's head-end node can set it; the two enabled-by-default rate limits of Section 4.4 (the marking-rate limit at the head-end node and the postcard-generation cap at every PBT-M-aware node) MUST bound the postcard load so that flag tampering cannot be amplified into a DoS; and the counters and rate-limited alarms (Section 5.4) SHOULD be used to detect abnormal marking or postcard rates indicative of tampering. Protection of the exported postcards and configuration against eavesdropping and modification can be provided by encrypting and/or authenticating them (Section 2). The security considerations of the MNA Sub-Stack in Section 11 of [RFC9994] also apply.

Beyond the P-flag manipulation, a compromised node inside the trust boundary can also forge postcard packets to the collector. The impact of such forged postcards on the telemetry system depends on the correlation and analytic algorithms in use. Postcards with non-existent or corrupted Flow-IDs cannot be correlated with legitimate path traces and will typically be discarded by the collector as anomalies, consuming collector bandwidth but failing to pollute valid flow records. On the other hand, postcards injected with valid Flow-IDs but falsified data cause telemetry poisoning, leading to inaccurate calculations, false path reconstruction, or spurious operational alarms. To mitigate the effect of forged postcard packets, collectors and data analytics systems SHOULD perform per-exporter rate limiting and data validation against the known topology.

A compromised node could also send crafted or malformed postcard packets to exploit collector vulnerabilities. Collectors that fail to enforce rigorous input validation could be vulnerable to buffer overflows, unhandled exceptions, or denial-of-service crashes. Therefore, collectors SHOULD treat postcard telemetry as untrusted input, enforce strict validation on all received fields, and safely handle or ignore malformed data and unknown extensions.

7. IANA Considerations

This document requests IANA to allocate one bit position (TBD1) for the PBT-M network action, with the description "PBT-M (Postcard-Based Telemetry with Packet Marking)", from the "Network Action Flags Without Ancillary Data" registry in the "MPLS Network Actions" registry group created by [RFC9994]. To preserve ECMP load-balancing, the allocated bit position needs to fall within a Format D LSE (word bits 24-31), i.e., outside the most significant 23 bits of the LSE, so that the mutable marking bit does not affect ECMP load-balancing whether or not the TC field is used in the hash (see Section 5.2 of [RFC9994]). The registration procedure for this range (bit positions 20-439) is IETF Review.

8. Acknowledgments

The authors thank Carlos Pignataro and Fung Lim for the OPSDIR review, Bruno Decraene and Gyan Mishra for the RTGDIR review, Christian Huitema for the SECDIR review, and Tony Li for the document shepherding.

9. References

9.1. Normative References

[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, , <https://www.rfc-editor.org/info/rfc2119>.
[RFC7011]
Claise, B., Ed., Trammell, B., Ed., and P. Aitken, "Specification of the IP Flow Information Export (IPFIX) Protocol for the Exchange of Flow Information", STD 77, RFC 7011, DOI 10.17487/RFC7011, , <https://www.rfc-editor.org/info/rfc7011>.
[RFC8174]
Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, , <https://www.rfc-editor.org/info/rfc8174>.
[RFC8476]
Tantsura, J., Chunduri, U., Aldrin, S., and P. Psenak, "Signaling Maximum SID Depth (MSD) Using OSPF", RFC 8476, DOI 10.17487/RFC8476, , <https://www.rfc-editor.org/info/rfc8476>.
[RFC8491]
Tantsura, J., Chunduri, U., Aldrin, S., and L. Ginsberg, "Signaling Maximum SID Depth (MSD) Using IS-IS", RFC 8491, DOI 10.17487/RFC8491, , <https://www.rfc-editor.org/info/rfc8491>.
[RFC9994]
Rajamanickam, J., Ed., Gandhi, R., Ed., Zigler, R., Song, H., and K. Kompella, "MPLS Network Action (MNA) Sub-Stack Specification Including In-Stack Network Actions and Data", RFC 9994, DOI 10.17487/RFC9994, , <https://www.rfc-editor.org/info/rfc9994>.

9.2. Informative References

[I-D.ietf-mpls-mna-ps-hdr]
Rajamanickam, J., Gandhi, R., Zigler, R., Dong, J., and J. Bhattacharya, "MPLS Network Action (MNA) Post-Stack Header Specification", Work in Progress, Internet-Draft, draft-ietf-mpls-mna-ps-hdr-19, , <https://datatracker.ietf.org/doc/html/draft-ietf-mpls-mna-ps-hdr-19>.
[I-D.ietf-opsawg-rfc5706bis]
Claise, B., Clarke, J., Farrel, A., Barguil, S., Pignataro, C., and R. Chen, "Guidelines for Considering Operations and Management in IETF Specifications", Work in Progress, Internet-Draft, draft-ietf-opsawg-rfc5706bis-09, , <https://datatracker.ietf.org/doc/html/draft-ietf-opsawg-rfc5706bis-09>.
[I-D.spiegel-ippm-ioam-rawexport]
Spiegel, M., Brockners, F., Bhandari, S., and R. Sivakolundu, "In-situ OAM raw data export with IPFIX", Work in Progress, Internet-Draft, draft-spiegel-ippm-ioam-rawexport-07, , <https://datatracker.ietf.org/doc/html/draft-spiegel-ippm-ioam-rawexport-07>.
[RFC0792]
Postel, J., "Internet Control Message Protocol", STD 5, RFC 792, DOI 10.17487/RFC792, , <https://www.rfc-editor.org/info/rfc792>.
[RFC5476]
Claise, B., Ed., Johnson, A., and J. Quittek, "Packet Sampling (PSAMP) Protocol Specifications", RFC 5476, DOI 10.17487/RFC5476, , <https://www.rfc-editor.org/info/rfc5476>.
[RFC5477]
Dietz, T., Claise, B., Aitken, P., Dressler, F., and G. Carle, "Information Model for Packet Sampling Exports", RFC 5477, DOI 10.17487/RFC5477, , <https://www.rfc-editor.org/info/rfc5477>.
[RFC7015]
Trammell, B., Wagner, A., and B. Claise, "Flow Aggregation for the IP Flow Information Export (IPFIX) Protocol", RFC 7015, DOI 10.17487/RFC7015, , <https://www.rfc-editor.org/info/rfc7015>.
[RFC9197]
Brockners, F., Ed., Bhandari, S., Ed., and T. Mizrahi, Ed., "Data Fields for In Situ Operations, Administration, and Maintenance (IOAM)", RFC 9197, DOI 10.17487/RFC9197, , <https://www.rfc-editor.org/info/rfc9197>.
[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, , <https://www.rfc-editor.org/info/rfc9259>.
[RFC9326]
Song, H., Gafni, B., Brockners, F., Bhandari, S., and T. Mizrahi, "In Situ Operations, Administration, and Maintenance (IOAM) Direct Exporting", RFC 9326, DOI 10.17487/RFC9326, , <https://www.rfc-editor.org/info/rfc9326>.
[RFC9789]
Andersson, L., Bryant, S., Bocci, M., and T. Li, "MPLS Network Actions (MNAs) Framework", RFC 9789, DOI 10.17487/RFC9789, , <https://www.rfc-editor.org/info/rfc9789>.

Authors' Addresses

Haoyu Song
Futurewei Technologies
United States of America
Giuseppe Fioccola
Huawei Technologies
Germany
Rakesh Gandhi
Cisco Systems
Canada