Summary

  • RFC 3637 says that several SONET-MIB counters used to represent Ethernet WIS performance are inhibited during severely errored or unavailable seconds, while the corresponding IEEE attributes are generalized nonresetable counters without that inhibition.
  • A flat code-violation counter can therefore coexist with a worsening state. Sound interpretation requires the interval, validity, defect, unavailable-time, layer, threshold and sampling evidence around the number.

Imagine an incident bridge watching a fifteen-minute error curve. The curve rises, the path deteriorates further, and then the curve becomes almost still. A dashboard caption might call that recovery. RFC 3637 supplies a less comforting explanation: some counters are deliberately not incremented once the second has crossed into a more serious classification.

The document concerns the Ethernet WAN Interface Sublayer, or WIS, used in 10GBASE-W physical layers. The WIS sits between the Physical Coding Sublayer and the Physical Medium Attachment sublayer, performing SONET/SDH-compatible framing and scrambling. That scope is precise. The RFC says its objects do not apply to other interface kinds, and it distinguishes a WIS port from Ethernet Line Terminating Equipment that uses related physical sublayers while behaving as SONET line equipment.

That precision matters because the management view is assembled from several object families. The Ethernet-like MIB describes the Ethernet-facing portion. The MAU-MIB identifies and controls the medium attachment. The Interfaces Group and inverted stack table locate the layers. The SONET-MIB supplies much of the section, line and path performance view. RFC 3637 adds a sparse augmentation for WIS-specific controls, trace messages and status conditions.

“Sparse augmentation” is not administrative trivia. The authors considered defining a fresh set of objects that exactly matched the IEEE WIS attributes. They rejected that route because deployed systems already used the SONET-MIB to manage SONET sublayers beneath ATM and PPP. Operational continuity won, but the RFC did not pretend the old and new meanings were identical. It itemised the differences.

The sharpest difference lies in time and inhibition. IEEE attributes such as code-violation counts were described as generalized nonresetable counters. The corresponding SNMP view uses current and 15-minute interval counters. Section code violations are inhibited during one-second intervals classified as severely errored. Line and path measurements have their own rules: various errored-second and code-violation counts are inhibited during unavailable seconds, and some code-violation counts are inhibited during either severely errored or unavailable seconds. Near-end and far-end tables do not all lose information in exactly the same way.

This produces a counterintuitive but legitimate observation. Suppose physical errors continue while the agent classifies the interval as severely errored or unavailable. A generalized IEEE counter would continue accumulating under the comparison described by the RFC. The imported SONET interval counter may stop incrementing. The quieter line is not falsified data. It is a truthful result under a different measurement contract.

The contract differs elsewhere too. IEEE severe-error thresholds were integers with one instance per interface. sonetSESthresholdSet is an enumerated value with one instance per network element; it controls all layers together and permits only selected values. At path level, IEEE criteria included Payload Label Mismatch and Loss of Codegroup Delineation conditions that the older SONET-MIB criteria did not. RFC 3637 adds WIS-specific current-status bits, but that does not erase the semantic difference in the imported counters.

Appendix A makes the transformation visible. An agent can poll latched status and rolling error registers once per second. It computes modular differences as 16- or 32-bit registers wrap, classifies the second, applies unavailable-time logic, decides which current counts may advance, rotates the current values into a history stack at each quarter-hour, and marks whether the interval is valid. A sample count outside 890 to 910 makes an interval invalid. Far-end counts acquire an additional invalidity condition when a near-end defect occurred.

The physical register is therefore only the first custody point. The agent adds time, rollover arithmetic, threshold comparison, state classification, inhibition, history rotation and validity. The MIB then exposes a projection. A collector samples that projection again. A dashboard chooses a rate or colour. A human turns the display into a claim about health. Each transition can preserve its own truth while discarding information required by the next decision.

Four layers should remain separate:

  1. The physical layer records defects and errors in WIS hardware.
  2. The collection layer polls, timestamps, handles rollover and classifies one-second windows.
  3. The projection layer increments or inhibits MIB counters and marks interval validity.
  4. The decision layer assigns operational meaning, responsibility and action.

The safest immediate practice is to join rather than rank. A code-violation series should travel with severe-error and unavailable-time state, interval-validity flags, elapsed time, sample coverage, threshold setting, the exact metric name and the relevant local or far-end status. If a chart is exported without those fields, the organisation has not merely compressed telemetry; it may have removed the explanation for why the chart became calm.

The same discipline applies to security. RFC 3637 identifies five read-write objects. Changing test-pattern modes or accumulated test errors can interfere with an in-progress pattern test. Changing transmitted section or path trace messages can raise connectivity alarms at the remote end. Even read access can expose performance and configuration. The RFC recommends SNMPv3 authentication and privacy and discourages earlier versions. A successful GET proves only that an authorised or unauthorised reader obtained a projected value; it does not prove the underlying collection or the operational conclusion.

None of this shows that a named product implemented the logic incorrectly, that a carrier suffered an outage, or that WIS remains prevalent. The standards-track status, IETF history, RFC Editor record, errata snapshot and IANA interface registry prove documentary facts. They do not prove deployment. RFC 3637 is valuable precisely because it lets us study a declared loss of equivalence without inventing an incident.

Sources