Performance Measurement (SRPM)¶

Segment Routing Performance Measurement (SRPM) provides real-time visibility into network path quality across SRv6 domains. By measuring delay, loss, and jitter on a per-path or per-flow basis, operators can enforce SLAs, optimize traffic engineering decisions, and detect degradation before it impacts services.

Why Performance Measurement Matters¶

In SRv6 networks, traffic follows explicit segment lists through the fabric. Knowing how well each path performs is critical for:

• SLA assurance -- Proving that contracted delay/loss targets are met
• TE optimization -- Feeding real-time metrics into SR Policy path selection
• Fault detection -- Identifying degraded links before they cause outages
• Capacity planning -- Understanding utilization patterns per path

Key Metrics¶

How SRPM Works¶

STAMP (RFC 8762)¶

Simple Two-Way Active Measurement Protocol is the modern standard for active probing in SRv6 networks. It replaces older TWAMP with a simpler, stateless design.

graph LR
    S[Sender PE] -->|"STAMP Probe (UDP)"| R[Reflector PE]
    R -->|"STAMP Reply + Timestamps"| S
    S --- M((PM Engine))
    M -->|"Delay, Loss, Jitter"| T[Telemetry Collector]

How it works:

1. Sender generates UDP probe packets with timestamps embedded in the SRv6 path
2. Reflector receives the probe, adds its own timestamps, and returns a reply
3. PM Engine computes one-way delay, round-trip delay, and loss from timestamp differences
4. Results are streamed to a telemetry collector for visualization and alerting

Alternate-Marking (RFC 9341 / RFC 9342)¶

Alternate-marking is a passive measurement technique that avoids injecting probe traffic entirely. Instead, it marks production packets with alternating color bits and counts them at ingress and egress.

By comparing ingress and egress counters for each color block, the system calculates exact packet loss without any probe overhead.

Color-Aware PM¶

SRv6 SR Policies can carry a color attribute that maps to an SLA class. Performance measurement sessions can be bound to specific colors, enabling per-SLA monitoring:

• Color 10 -- Low-latency path (delay < 5ms)
• Color 20 -- Best-effort path
• Color 30 -- High-bandwidth path

This allows operators to measure whether each SLA class is meeting its target independently.

Measurement Approaches Compared¶

SRv6 PM Integration¶

PM with SR Policies¶

Performance measurement integrates directly with SR Policies. Each candidate path in an SR Policy can have PM enabled:

segment-routing
 traffic-eng
  policy POLICY-1
   color 10 end-point fc00:0:2::1
   candidate-paths
    preference 100
     explicit segment-list SL-LOW-LATENCY
    !
   performance-measurement
    liveness-detection
    delay-measurement
    loss-measurement
   !

Anomaly Thresholds and Metric Fallback¶

When a PM session detects degradation, the router can advertise an anomaly and automatically raise the IS-IS metric on the affected interface to steer traffic away. The key to doing this without flapping is the hysteresis band — a pair of thresholds that separate the "enter anomaly" and "exit anomaly" conditions.

performance-measurement
 delay-profile interfaces default
  advertisement
   anomaly-loss
    upper-bound 30      ← anomaly declared when loss exceeds 30%
    lower-bound 20      ← anomaly cleared only when loss drops below 20%
   !
  !
 !

router isis 1
 interface GigabitEthernet0/2/0/1
  point-to-point
  address-family ipv4 unicast
   metric fallback anomaly loss increment 500
  !
  address-family ipv6 unicast
   metric fallback anomaly loss increment 500
  !
 !

How the hysteresis band works:

graph LR
    A["Loss < 20%<br/>(Normal)"] -->|"Loss rises above 30%"| B["Loss > 30%<br/>(Anomaly declared)"]
    B -->|"Metric += 500<br/>advertised in IS-IS"| C["Traffic rerouted"]
    C -->|"Loss drops below 20%"| D["Anomaly cleared<br/>Metric restored"]
    D --> A
    B -->|"Loss between 20–30%<br/>no state change"| B
    style B fill:#b71c1c26,stroke:#b71c1c
    style D fill:#1b5e2026,stroke:#1b5e20

The gap between upper-bound and lower-bound is the dead band (also called hysteresis zone). Loss oscillating in that range — say between 21% and 29% — does not trigger any state change, preventing metric flapping. Without hysteresis, a link hovering at exactly the threshold would cause the IS-IS metric to oscillate continuously.

Streaming PM Data¶

PM results are exported via streaming telemetry for real-time dashboards and alerting. See Telemetry & Monitoring for collector setup.

Key YANG paths for PM data:

Vendor Support¶

Huawei is not in this table because we could not verify release numbers for it: the NetEngine 8000 V800R025 STAMP and TWAMP Light chapters do not cover SRv6, and Huawei documents SRv6 path measurement as IFIT (alternate marking). For SRv6-specific performance measurement on each platform, see the SRv6 Support Matrix.

Benefits¶

Further Reading¶

• OAM & Troubleshooting -- Ping, traceroute, and OAM tools for SRv6
• Telemetry & Monitoring -- Streaming telemetry and dashboards for SRv6
• Traffic Engineering -- How PM feeds into SR Policy optimization
• Flex-Algorithm -- Constraint-based path computation using PM metrics
• TI-LFA -- Fast reroute protection complementing PM liveness detection

References¶

1. RFC 8762 - Simple Two-Way Active Measurement Protocol (STAMP) -- Defines the STAMP protocol for active delay and loss measurement
2. RFC 8972 - STAMP Optional Extensions -- Extensions to STAMP for additional measurement capabilities
3. RFC 9341 - Alternate-Marking Method -- Defines the alternate-marking technique for passive loss measurement
4. RFC 9342 - Clustered Alternate-Marking Method -- Extends alternate-marking for clustered deployments
5. RFC 9256 - SR Policy Architecture -- SR Policy framework that integrates with PM