Summary

  • RFC 1304 counted only unerrored Level 3 receipts in its receive objects and defined sipL3Errors around discarded protocol-processing or bit errors, explicitly excluding addressing-related errors.
  • The memo assembled other semantic error counters separately and warned that subscription violations could cause additional discards recorded in another MIB; no one object was an exhaustive loss ledger.
  • Its error table retained only the latest occurrence per interface and error type, while RFC 1694 later moved several counts into ifTable; scope, time and source therefore had to travel with every operational claim.

The dangerous comfort of a total

RFC 1304 arrived in February 1992 as a Proposed Standard for managing the SIP interface type—the SMDS Interface Protocol—through SNMP. It described a layered access stack: SIP Level 3 above SIP Level 2, a physical-layer convergence procedure, and DS1 or DS3 beneath. The document did not pretend that one table owned the entire stack. Existing MIBs covered the transmission interfaces, and another specification covered SMDS subscription and Subscriber-Network Interface configuration.

That architecture matters more than the age of SMDS. A manager rarely encounters “the network” as one record. It encounters counters whose names sound comprehensive because their denominators have been hidden by the display. RFC 1304 exposed those denominators in prose.

The receive counters for individually addressed and group-addressed Level 3 PDUs included only unerrored packets. They were not ingress totals before validation. A packet excluded from those objects had not vanished from reality; it had fallen outside that particular population.

One discarded packet could belong to several ledgers

The object named sipL3Errors counted Level 3 PDUs received from the remote system, discovered to contain protocol-processing or bit errors, and discarded. Its description excluded addressing-related errors. RFC 1304 then explained that the total Level 3 PDU error calculation combined a syntactic component with semantic-service components: unrecognized individual destinations, unrecognized group destinations and invalid SMDS address types.

Even that assembly did not close the world. The memo stated that public SMDS networks could discard PDUs for subscription violations and directed the reader to the SMDS Subscription MIB. Destination Address Screening violations were expressly absent from the unrecognized-destination counters.

The result is not a contradiction. Each object answered a bounded question. The problem begins when a collector promotes one answer into “all loss” without retaining the exclusions. A zero protocol-error counter can coexist with policy screening. A rising unrecognized-address counter can coexist with clean framing. An unerrored receive count can rise while an application still fails to use the data.

The log was a last witness, not an archive

RFC 1304 also created a table for twelve syntactical Level 3 error types. The catalogue is precise: malformed address fields, impossible sizes, bad header-extension positions or lengths, tag mismatches, length disagreement and an expired Message Receive Interval. For each interface and type, the table retained a rejected source address, rejected destination address and timestamp.

But the table described the latest occurrence. A new disagreement of the same type displaced the earlier one. It did not count a chronology, preserve every affected packet or record a repair. The timestamp itself used sysUpTime, tying interpretation to the managed system’s time base rather than a civil-time incident clock. When the timestamp was zero, the row contained no valid information.

That zero is disciplined ignorance. It does not certify that no subscription discard, application failure or unrepresented event occurred. It says only that this latest-occurrence slot has no valid record under its own contract.

A placeholder was not a capability

The MIB tree included an SMDS carrier-selection group, but RFC 1304 called it a placeholder. Naming a branch reserved a coordinate; it did not populate the branch with operational objects. This is the same evidentiary mistake in another form: an address in a schema is not a running feature.

The IP-over-SMDS application table supplied a richer example. It could associate an IP address with an individual SMDS address, the group address for a Logical IP Subnetwork and the destination used for ARP requests. One SMDS address could serve multiple IP addresses, and one Subscriber-Network Interface could participate in multiple logical subnetworks. A single row was therefore not a unique claim about one physical customer, one network or one service.

Those objects were defined read-only. RFC 1304 nevertheless allowed an agent, at its discretion, to expose them as read-write so an appropriately authorized management station could change logical-subnetwork addressing. Definition, implementation choice, authorization, requested change and resulting configuration were five different facts. The MIB did not collapse them merely because they shared an object name.

The measurement address moved

RFC 1694 replaced RFC 1304 in 1994 with an SMIv2-compliant module and described the definitions as semantically identical. It also deprecated several SIP counters and directed managers toward corresponding objects in the MIB-II interfaces table. Intended meaning could survive while the canonical query location changed.

That creates a historical join problem. A collector moving from the old object to ifTable needed to know when the source changed, whether both were sampled during overlap, whether their populations were truly comparable and whether counter epochs were preserved. The replacement document did not prove that any product or archive performed that migration correctly.

A smooth chart can conceal a broken provenance chain. Conversely, a visible step can be a measurement migration rather than a network event. Without the object identifier, module version, interface identity, sampling interval and reset context, a number loses the evidence needed to interpret it.

What the MIB could not decide

RFC 1304 made real operational facts observable. It did not make them universal. It said nothing about a named deployment and reported no live values. It did not establish that an agent implemented every mandatory group correctly, that a management station fetched the right rows, or that a service interruption followed from any counter.

The memo’s security section simply said security issues were not discussed. That omission cannot be repaired by assuming the management channel was authenticated or the data trustworthy. Nor can the existence of a writable option establish that a change was authorized.

The strongest historical lesson is modest: the early Internet management architecture knew that one surface should not pretend to own every cause. Protocol faults, address semantics, subscription policy, configuration and application results were separated. A modern dashboard becomes less reliable, not more, when it erases those boundaries behind one impressive total.

Sources