Summary

  • RFC 2127 represented physical ISDN access, D-channel signaling, individual B channels and upper interfaces as separate ifTable records joined through ifStack; that graph was an asserted management topology, not proof of an end-to-end circuit.
  • A hyperchannel could occupy several B channels while counting as one signaling call. Its bearer rows repeated the same call values, so neither row count nor channel count established the number of calls or usable delivered capacity.
  • Removing the local ifStack binding was the specified disconnect action. It did not, by itself, prove remote teardown, final charging, payload delivery or the application result.

Imagine a router display with one PPP interface above four B-channel interfaces. All four bearer rows show the same peer, origin, setup time and charging value. The signaling counter advances once. Which number describes the connection: one, four or five?

In March 1997, RFC 2127 answered by refusing to force those numbers into one object. An ISDN interface usually contained a physical Basic Rate or Primary Rate interface, a D channel for signaling, several B channels for bearer traffic and an encapsulation layer such as PPP. The specification gave the layers separate ifTable identities and joined them with the Interfaces MIB's ifStack relationships. The resulting graph made control points visible. It did not make them interchangeable.

One call, several rows

The physical interface described the access. A D channel could have a LAPD data-link entity and a network-layer signaling entity. Every B channel received its own interface row and one isdnBearerTable entry. An upper interface represented the peer-facing encapsulation. The stack recorded which layer ran over which lower layer.

This was a disciplined decomposition. A manager could observe a physical port separately from signaling, see whether a particular B channel was idle, connecting, validating an incoming call or active, and bind an upper interface only to the bearer then in use. The companion RFC 2128 Dial Control MIB held generic peer configuration, active-call and call-history information so the ISDN-specific MIB did not claim to own every demand-access meaning.

The separation also bounded the evidence. A row proved that an agent exposed an object with defined semantics. An ifStack entry asserted a local relationship between two conceptual interfaces. Neither statement certified the remote switch, the far endpoint, the application session or the traffic actually delivered.

Hyperchannels broke row-count intuition

RFC 2127's sharpest example was the hyperchannel. One call could use several B channels. There was one encapsulation-layer interface but several lower bearer interfaces. An implementation could represent the fan-out through an optional DS0Bundle or through several ifStack entries sharing one higher-layer index and using different lower-layer indexes.

Every B-channel row associated with the hyperchannel carried identical values for the peer address and subaddress, call origin, information type, multirate flag, setup and connect times, and charged units. Yet the isdnSignalingStatsTable treated the entire hyperchannel as one call regardless of how many B channels it occupied.

That makes three common inferences unsafe. Four bearer rows did not mean four calls. One call did not mean one bearer channel. Four allocated channels did not prove four channels' worth of usable payload reached the peer. The model exposed allocation and grouping; capacity and delivery required their own observations.

The later RFC 2494 DS0 and DS0Bundle MIB formalized bundle representation in 1999. It is useful later context, not evidence that every 1997 implementation chose or correctly deployed that option.

The stack was an asserted topology

RFC 2127 depended on the general model in RFC 1573: each sublayer corresponded to a conceptual ifTable row, while ifStackTable recorded which sublayer sat on top of another. That scheme was strong precisely because it avoided pretending that one interface counter could describe every layer.

It also meant that a topology view had an owner. The local agent created and reported the rows. A correct graph could show the relationships the implementation believed active. A stale or incomplete graph could mislead without violating the syntax. The RFC did not provide an independent end-to-end witness that every represented relationship carried useful service at that instant.

D-channel backup and Non-Facility Associated Signaling trunks made the point again. Several LAPD interfaces could exist while only one was active and the others dormant; one signaling entity could also sit above one LAPD interface and several physical interfaces. The graph described role and attachment, not a single cable with a single truth value.

Disconnect was a local mutation

The bearer-group definition stated how to disconnect an active call: remove the ifStack row binding the higher-layer interface to the B-channel interface. This was an executable management action, not merely a status label.

But its receipt was narrow. A successful local removal could show that the agent accepted the requested topology change. It did not necessarily prove that the far end cleared, the network completed every release exchange, the charging system closed its record, queued payload disappeared, or the application interpreted the event correctly. Those conclusions belonged to other layers.

This distinction is not a criticism of the MIB. It is the reason the MIB was useful. By naming the local action precisely, it made the missing downstream receipts easier to identify.

Fields with bounded meanings

Several objects warned against treating a management value as a universal fact.

The peer address described the current or last call. Its exact format could depend on the switch or PBX, and the value could be empty when unavailable. A populated string was therefore neither guaranteed current nor automatically normalized, authenticated or comparable across equipment.

isdnBearerChargedUnits also referred to the current or last connection. It was zero for incoming calls or when the switch supplied no charging information. Zero could mean no reported charge, not necessarily no economic cost. The signaling table's accumulated charging counter was narrower again: it counted charging units applying to the local interface, including originated calls and reverse charging.

Dial-up and leased B channels had different control surfaces. A dial-up bearer was controlled by an associated signaling channel. A leased bearer had no signaling-channel control, and an entirely leased Primary Rate interface could be handled by the DS1/E1 MIB without ISDN-specific rows. Asking the signaling table to explain a leased bearer would be asking the wrong layer.

Speech, 3.1 kHz audio and 7 kHz audio labels were signaling capabilities, not complete descriptions of treatment. RFC 2127 called them signaling artifices and left the network to decide what to do within the promised service. A speech bearer might be compressed or otherwise processed while preserving voice quality; it did not guarantee that a modem would work.

A standard with an explicit blank space

The official RFC record and IETF history identify the document as a Proposed Standard produced by the ISDN MIB working group. The RFC Editor errata record contains one verified technical correction to a conformance object identifier and one rejected report. Neither supplies operational deployment evidence.

The security section was one sentence: security issues were not discussed. That absence must remain an absence. It is not permission to infer confidentiality, authentication, safe write access or resistance to a hostile manager.

The durable achievement of RFC 2127 was not that it put an entire call inside one reliable status light. It did the opposite. It provided different objects for different claims and let their relationships be inspected. The operator's task was to keep those claims from collapsing back into one another.

Sources and limits

The official sources establish the 1997 model and its object semantics, not product prevalence, a named deployment, a measured call, billing accuracy or present-day interoperability. Running-Code Primacy, Minimum Initial Specification and Reality Layers are later Lu Heng essays used here only as an inference discipline: specification, local executable state and operational outcome are separate evidence layers. They are not evidence of the RFC authors' intention.