Tech-invite3GPPspaceIETFspace
96959493929190898887868584838281807978777675747372717069686766656463626160595857565554535251504948474645444342414039383736353433323130292827262524232221201918171615141312111009080706050403020100
in Index   Prev   Next

RFC 2863

The Interfaces Group MIB

Pages: 69
Draft Standard
Errata
Obsoletes:  2233
Updated by:  8892
Part 1 of 3 – Pages 1 to 26
None   None   Next

Top   ToC   RFC2863 - Page 1
Network Working Group                                      K. McCloghrie
Request for Comments: 2863                                 Cisco Systems
Obsoletes: 2233                                            F. Kastenholz
Category: Standards Track                                 Argon Networks
                                                               June 2000


                        The Interfaces Group MIB

Status of this Memo

   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

Copyright Notice

   Copyright (C) The Internet Society (2000).  All Rights Reserved.

Table of Contents

1 Introduction ................................................. 2 2 The SNMP Network Management Framework ........................ 2 3 Experience with the Interfaces Group ......................... 3 3.1 Clarifications/Revisions ................................... 4 3.1.1 Interface Sub-Layers ..................................... 4 3.1.2 Guidance on Defining Sub-layers .......................... 7 3.1.3 Virtual Circuits ......................................... 8 3.1.4 Bit, Character, and Fixed-Length Interfaces .............. 8 3.1.5 Interface Numbering ...................................... 10 3.1.6 Counter Size ............................................. 14 3.1.7 Interface Speed .......................................... 16 3.1.8 Multicast/Broadcast Counters ............................. 17 3.1.9 Trap Enable .............................................. 17 3.1.10 Addition of New ifType values ........................... 18 3.1.11 InterfaceIndex Textual Convention ....................... 18 3.1.12 New states for IfOperStatus ............................. 18 3.1.13 IfAdminStatus and IfOperStatus .......................... 19 3.1.14 IfOperStatus in an Interface Stack ...................... 21 3.1.15 Traps ................................................... 21 3.1.16 ifSpecific .............................................. 23 3.1.17 Creation/Deletion of Interfaces ......................... 23 3.1.18 All Values Must be Known ................................ 24 4 Media-Specific MIB Applicability ............................. 24 5 Overview ..................................................... 25 6 Interfaces Group Definitions ................................. 26
Top   ToC   RFC2863 - Page 2
   7 Acknowledgements .............................................   64
   8 References ...................................................   64
   9 Security Considerations ......................................   66
   10 Authors' Addresses ..........................................   67
   11 Changes from RFC 2233 .......................................   67
   12 Notice on Intellectual Property .............................   68
   13 Full Copyright Statement ....................................   69

1. Introduction

This memo defines a portion of the Management Information Base (MIB) for use with network management protocols in the Internet community. In particular, it describes managed objects used for managing Network Interfaces. This memo discusses the 'interfaces' group of MIB-II [17], especially the experience gained from the definition of numerous media-specific MIB modules for use in conjunction with the ' interfaces' group for managing various sub-layers beneath the internetwork-layer. It specifies clarifications to, and extensions of, the architectural issues within the MIB-II model of the ' interfaces' group. This memo obsoletes RFC 2233, the previous version of the Interfaces Group MIB. The key words "MUST" and "MUST NOT" in this document are to be interpreted as described in RFC 2119 [16].

2. The SNMP Network Management Framework

The SNMP Management Framework presently consists of five major components: o An overall architecture, described in RFC 2571 [1]. o Mechanisms for describing and naming objects and events for the purpose of management. The first version of this Structure of Management Information (SMI) is called SMIv1 and described in STD 16, RFC 1155 [2], STD 16, RFC 1212 [3] and RFC 1215 [4]. The second version, called SMIv2, is described in STD 58, which consists of RFC 2578 [5], RFC 2579 [6] and RFC 2580 [7]. o Message protocols for transferring management information. The first version of the SNMP message protocol is called SNMPv1 and described in STD 15, RFC 1157 [8]. A second version of the SNMP message protocol, which is not an Internet standards track protocol, is called SNMPv2c and described in RFC 1901 [9] and RFC 1906 [10]. The third version of the message protocol is called SNMPv3 and described in RFC 1906 [10], RFC 2572 [11] and RFC 2574 [12].
Top   ToC   RFC2863 - Page 3
      o  Protocol operations for accessing management information.  The
         first set of protocol operations and associated PDU formats is
         described in STD 15, RFC 1157 [8].  A second set of protocol
         operations and associated PDU formats is described in RFC 1905
         [13].

      o  A set of fundamental applications described in RFC 2573 [14]
         and the view-based access control mechanism described in RFC
         2575 [15].

   A more detailed introduction to the current SNMP Management Framework
   can be found in RFC 2570 [22].

   Managed objects are accessed via a virtual information store, termed
   the Management Information Base or MIB.  Objects in the MIB are
   defined using the mechanisms defined in the SMI.

   This memo specifies a MIB module that is compliant to the SMIv2.  A
   MIB conforming to the SMIv1 can be produced through the appropriate
   translations.  The resulting translated MIB must be semantically
   equivalent, except where objects or events are omitted because no
   translation is possible (e.g., use of Counter64).  Some machine
   readable information in SMIv2 will be converted into textual
   descriptions in SMIv1 during the translation process.  However, this
   loss of machine readable information is not considered to change the
   semantics of the MIB.

3. Experience with the Interfaces Group

One of the strengths of internetwork-layer protocols such as IP [18] is that they are designed to run over any network interface. In achieving this, IP considers any and all protocols it runs over as a single "network interface" layer. A similar view is taken by other internetwork-layer protocols. This concept is represented in MIB-II by the 'interfaces' group which defines a generic set of managed objects such that any network interface can be managed in an interface-independent manner through these managed objects. The ' interfaces' group provides the means for additional managed objects specific to particular types of network interface (e.g., a specific medium such as Ethernet) to be defined as extensions to the ' interfaces' group for media-specific management. Since the standardization of MIB-II, many such media-specific MIB modules have been defined. Experience in defining these media-specific MIB modules has shown that the model defined by MIB-II is too simplistic and/or static for some types of media-specific management. As a result, some of these media-specific MIB modules assume an evolution or loosening of the
Top   ToC   RFC2863 - Page 4
   model.  This memo documents and standardizes that evolution of the
   model and fills in the gaps caused by that evolution.  This memo also
   incorporates the interfaces group extensions documented in RFC 1229
   [19].

3.1. Clarifications/Revisions

There are several areas for which experience has indicated that clarification, revision, or extension of the model would be helpful. The following sections discuss the changes in the interfaces group adopted by this memo in each of these areas. In some sections, one or more paragraphs contain discussion of rejected alternatives to the model adopted in this memo. Readers not familiar with the MIB-II model and not interested in the rationale behind the new model may want to skip these paragraphs.

3.1.1. Interface Sub-Layers

Experience in defining media-specific management information has shown the need to distinguish between the multiple sub-layers beneath the internetwork-layer. In addition, there is a need to manage these sub-layers in devices (e.g., MAC-layer bridges) which are unaware of which, if any, internetwork protocols run over these sub-layers. As such, a model of having a single conceptual row in the interfaces table (MIB-II's ifTable) represent a whole interface underneath the internetwork-layer, and having a single associated media-specific MIB module (referenced via the ifType object) is too simplistic. A further problem arises with the value of the ifType object which has enumerated values for each type of interface. Consider, for example, an interface with PPP running over an HDLC link which uses a RS232-like connector. Each of these sub-layers has its own media-specific MIB module. If all of this is represented by a single conceptual row in the ifTable, then an enumerated value for ifType is needed for that specific combination which maps to the specific combination of media-specific MIBs. Furthermore, such a model still lacks a method to describe the relationship of all the sub-layers of the MIB stack. An associated problem is that of upward and downward multiplexing of the sub-layers. An example of upward multiplexing is MLP (Multi- Link-Procedure) which provides load-sharing over several serial lines by appearing as a single point-to-point link to the sub-layer(s) above. An example of downward multiplexing would be several instances of PPP, each framed within a separate X.25 virtual circuit,
Top   ToC   RFC2863 - Page 5
   all of which run over one fractional T1 channel, concurrently with
   other uses of the T1 link.  The MIB structure must allow these sorts
   of relationships to be described.

   Several solutions for representing multiple sub-layers were rejected.
   One was to retain the concept of one conceptual row for all the sub-
   layers of an interface and have each media-specific MIB module
   identify its "superior" and "subordinate" sub-layers through OBJECT
   IDENTIFIER "pointers".  This scheme would have several drawbacks: the
   superior/subordinate pointers would be contained in the media-
   specific MIB modules; thus, a manager could not learn the structure
   of an interface without inspecting multiple pointers in different MIB
   modules; this would be overly complex and only possible if the
   manager had knowledge of all the relevant media-specific MIB modules;
   MIB modules would all need to be retrofitted with these new
   "pointers"; this scheme would not adequately address the problem of
   upward and downward multiplexing; and finally, enumerated values of
   ifType would be needed for each combination of sub-layers.  Another
   rejected solution also retained the concept of one conceptual row for
   all the sub-layers of an interface but had a new separate MIB table
   to identify the "superior" and "subordinate" sub-layers and to
   contain OBJECT IDENTIFIER "pointers" to the media-specific MIB module
   for each sub-layer.  Effectively, one conceptual row in the ifTable
   would represent each combination of sub-layers between the
   internetwork-layer and the wire.  While this scheme has fewer
   drawbacks, it still would not support downward multiplexing, such as
   PPP over MLP: observe that MLP makes two (or more) serial lines
   appear to the layers above as a single physical interface, and thus
   PPP over MLP should appear to the internetwork-layer as a single
   interface; in contrast, this scheme would result in two (or more)
   conceptual rows in the ifTable, both of which the internetwork-layer
   would run over.  This scheme would also require enumerated values of
   ifType for each combination of sub-layers.

   The solution adopted by this memo is to have an individual conceptual
   row in the ifTable to represent each sub-layer, and have a new
   separate MIB table (the ifStackTable, see section 6 below) to
   identify the "superior" and "subordinate" sub-layers through INTEGER
   "pointers" to the appropriate conceptual rows in the ifTable.  This
   solution supports both upward and downward multiplexing, allows the
   IANAifType to Media-Specific MIB mapping to identify the media-
   specific MIB module for that sub-layer, such that the new table need
   only be referenced to obtain information about layering, and it only
   requires enumerated values of ifType for each sub-layer, not for
   combinations of them.  However, it does require that the descriptions
   of some objects in the ifTable (specifically, ifType, ifPhysAddress,
   ifInUcastPkts, and ifOutUcastPkts) be generalized so as to apply to
   any sub-layer (rather than only to a sub-layer immediately beneath
Top   ToC   RFC2863 - Page 6
   the network layer as previously), plus some (specifically, ifSpeed)
   which need to have appropriate values identified for use when a
   generalized definition does not apply to a particular sub-layer.

   In addition, this adopted solution makes no requirement that a
   device, in which a sub-layer is instrumented by a conceptual row of
   the ifTable, be aware of whether an internetwork protocol runs on top
   of (i.e., at some layer above) that sub-layer.  In fact, the counters
   of packets received on an interface are defined as counting the
   number "delivered to a higher-layer protocol".  This meaning of
   "higher-layer" includes:

   (1)   Delivery to a forwarding module which accepts
         packets/frames/octets and forwards them on at the same protocol
         layer.  For example, for the purposes of this definition, the
         forwarding module of a MAC-layer bridge is considered as a
         "higher-layer" to the MAC-layer of each port on the bridge.

   (2)   Delivery to a higher sub-layer within a interface stack.  For
         example, for the purposes of this definition, if a PPP module
         operated directly over a serial interface, the PPP module would
         be considered the higher sub-layer to the serial interface.

   (3)   Delivery to a higher protocol layer which does not do packet
         forwarding for sub-layers that are "at the top of" the
         interface stack.  For example, for the purposes of this
         definition, the local IP module would be considered the higher
         layer to a SLIP serial interface.

   Similarly, for output, the counters of packets transmitted out an
   interface are defined as counting the number "that higher-level
   protocols requested to be transmitted".  This meaning of "higher-
   layer" includes:

   (1)   A forwarding module, at the same protocol layer, which
         transmits packets/frames/octets that were received on an
         different interface.  For example, for the purposes of this
         definition, the forwarding module of a MAC-layer bridge is
         considered as a "higher-layer" to the MAC-layer of each port on
         the bridge.

   (2)   The next higher sub-layer within an interface stack.  For
         example, for the purposes of this definition, if a PPP module
         operated directly over a serial interface, the PPP module would
         be a "higher layer" to the serial interface.
Top   ToC   RFC2863 - Page 7
   (3)   For sub-layers that are "at the top of" the interface stack, a
         higher element in the network protocol stack.  For example, for
         the purposes of this definition, the local IP module would be
         considered the higher layer to an Ethernet interface.

3.1.2. Guidance on Defining Sub-layers

The designer of a media-specific MIB must decide whether to divide the interface into sub-layers or not, and if so, how to make the divisions. The following guidance is offered to assist the media- specific MIB designer in these decisions. In general, the number of entries in the ifTable should be kept to the minimum required for network management. In particular, a group of related interfaces should be treated as a single interface with one entry in the ifTable providing that: (1) None of the group of interfaces performs multiplexing for any other interface in the agent, (2) There is a meaningful and useful way for all of the ifTable's information (e.g., the counters, and the status variables), and all of the ifTable's capabilities (e.g., write access to ifAdminStatus), to apply to the group of interfaces as a whole. Under these circumstances, there should be one entry in the ifTable for such a group of interfaces, and any internal structure which needs to be represented to network management should be captured in a MIB module specific to the particular type of interface. Note that application of bullet 2 above to the ifTable's ifType object requires that there is a meaningful media-specific MIB and a meaningful ifType value which apply to the group of interfaces as a whole. For example, it is not appropriate to treat an HDLC sub-layer and an RS-232 sub-layer as a single ifTable entry when the media- specific MIBs and the ifType values for HDLC and RS-232 are separate (rather than combined). Subject to the above, it is appropriate to assign an ifIndex value to any interface that can occur in an interface stack (in the ifStackTable) where the bottom of the stack is a physical interface (ifConnectorPresent has the value 'true') and there is a layer-3 or other application that "points down" to the top of this stack. An example of an application that points down to the top of the stack is the Character MIB [21].
Top   ToC   RFC2863 - Page 8
   Note that the sub-layers of an interface on one device will sometimes
   be different from the sub-layers of the interconnected interface of
   another device; for example, for a frame-relay DTE interface
   connected a frameRelayService interface, the inter-connected DTE and
   DCE interfaces have different ifType values and media-specific MIBs.

   These guidelines are just that, guidelines.  The designer of a
   media-specific MIB is free to lay out the MIB in whatever SMI
   conformant manner is desired.  However, in doing so, the media-
   specific MIB MUST completely specify the sub-layering model used for
   the MIB, and provide the assumptions, reasoning, and rationale used
   to develop that model.

3.1.3. Virtual Circuits

Several of the sub-layers for which media-specific MIB modules have been defined are connection oriented (e.g., Frame Relay, X.25). Experience has shown that each effort to define such a MIB module revisits the question of whether separate conceptual rows in the ifTable are needed for each virtual circuit. Most, if not all, of these efforts to date have decided to have all virtual circuits reference a single conceptual row in the ifTable. This memo strongly recommends that connection-oriented sub-layers do not have a conceptual row in the ifTable for each virtual circuit. This avoids the proliferation of conceptual rows, especially those which have considerable redundant information. (Note, as a comparison, that connection-less sub-layers do not have conceptual rows for each remote address.) There may, however, be circumstances under which it is appropriate for a virtual circuit of a connection- oriented sub-layer to have its own conceptual row in the ifTable; an example of this might be PPP over an X.25 virtual circuit. The MIB in section 6 of this memo supports such circumstances. If a media-specific MIB wishes to assign an entry in the ifTable to each virtual circuit, the MIB designer must present the rationale for this decision in the media-specific MIB's specification.

3.1.4. Bit, Character, and Fixed-Length Interfaces

RS-232 is an example of a character-oriented sub-layer over which (e.g., through use of PPP) IP datagrams can be sent. Due to the packet-based nature of many of the objects in the ifTable, experience has shown that it is not appropriate to have a character-oriented sub-layer represented by a whole conceptual row in the ifTable.
Top   ToC   RFC2863 - Page 9
   Experience has also shown that it is sometimes desirable to have some
   management information for bit-oriented interfaces, which are
   similarly difficult to represent by a whole conceptual row in the
   ifTable.  For example, to manage the channels of a DS1 circuit, where
   only some of the channels are carrying packet-based data.

   A further complication is that some subnetwork technologies transmit
   data in fixed length transmission units.  One example of such a
   technology is cell relay, and in particular Asynchronous Transfer
   Mode (ATM), which transmits data in fixed-length cells.  Representing
   such a interface as a packet-based interface produces redundant
   objects if the relationship between the number of packets and the
   number of octets in either direction is fixed by the size of the
   transmission unit (e.g., the size of a cell).

   About half the objects in the ifTable are applicable to every type of
   interface: packet-oriented, character-oriented, and bit-oriented.  Of
   the other half, two are applicable to both character-oriented and
   packet-oriented interfaces, and the rest are applicable only to
   packet-oriented interfaces.  Thus, while it is desirable for
   consistency to be able to represent any/all types of interfaces in
   the ifTable, it is not possible to implement the full ifTable for
   bit- and character-oriented sub-layers.

   A rejected solution to this problem would be to split the ifTable
   into two (or more) new MIB tables, one of which would contain objects
   that are relevant only to packet-oriented interfaces (e.g., PPP), and
   another that may be used by all interfaces.  This is highly
   undesirable since it would require changes in every agent
   implementing the ifTable (i.e., just about every existing SNMP
   agent).

   The solution adopted in this memo builds upon the fact that
   compliance statements in SMIv2 (in contrast to SMIv1) refer to object
   groups, where object groups are explicitly defined by listing the
   objects they contain.  Thus, with SMIv2, multiple compliance
   statements can be specified, one for all interfaces and additional
   ones for specific types of interfaces.  The separate compliance
   statements can be based on separate object groups, where the object
   group for all interfaces can contain only those objects from the
   ifTable which are appropriate for every type of interfaces.  Using
   this solution, every sub-layer can have its own conceptual row in the
   ifTable.

   Thus, section 6 of this memo contains definitions of the objects of
   the existing 'interfaces' group of MIB-II, in a manner which is both
   SNMPv2-compliant and semantically-equivalent to the existing MIB-II
   definitions.  With equivalent semantics, and with the BER ("on the
Top   ToC   RFC2863 - Page 10
   wire") encodings unchanged, these definitions retain the same OBJECT
   IDENTIFIER values as assigned by MIB-II.  Thus, in general, no
   rewrite of existing agents which conform to MIB-II and the
   ifExtensions MIB is required.

   In addition, this memo defines several object groups for the purposes
   of defining which objects apply to which types of interface:

   (1)   the ifGeneralInformationGroup.  This group contains those
         objects applicable to all types of network interfaces,
         including bit-oriented interfaces.

   (2)   the ifPacketGroup.  This group contains those objects
         applicable to packet-oriented network interfaces.

   (3)   the ifFixedLengthGroup.  This group contains the objects
         applicable not only to character-oriented interfaces, such as
         RS-232, but also to those subnetwork technologies, such as
         cell-relay/ATM, which transmit data in fixed length
         transmission units.  As well as the octet counters, there are
         also a few other counters (e.g., the error counters) which are
         useful for this type of interface, but are currently defined as
         being packet-oriented.  To accommodate this, the definitions of
         these counters are generalized to apply to character-oriented
         interfaces and fixed-length-transmission interfaces.

   It should be noted that the octet counters in the ifTable aggregate
   octet counts for unicast and non-unicast packets into a single octet
   counter per direction (received/transmitted).  Thus, with the above
   definition of fixed-length-transmission interfaces, where such
   interfaces which support non-unicast packets, separate counts of
   unicast and multicast/broadcast transmissions can only be maintained
   in a media-specific MIB module.

3.1.5. Interface Numbering

MIB-II defines an object, ifNumber, whose value represents: "The number of network interfaces (regardless of their current state) present on this system." Each interface is identified by a unique value of the ifIndex object, and the description of ifIndex constrains its value as follows: "Its value ranges between 1 and the value of ifNumber. The value for each interface must remain constant at least from one re-initialization of the entity's network management system to the next re-initialization."
Top   ToC   RFC2863 - Page 11
   This constancy requirement on the value of ifIndex for a particular
   interface is vital for efficient management.  However, an increasing
   number of devices allow for the dynamic addition/removal of network
   interfaces.  One example of this is a dynamic ability to configure
   the use of SLIP/PPP over a character-oriented port.  For such dynamic
   additions/removals, the combination of the constancy requirement and
   the restriction that the value of ifIndex is less than ifNumber is
   problematic.

   Redefining ifNumber to be the largest value of ifIndex was rejected
   since it would not help.  Such a re-definition would require ifNumber
   to be deprecated and the utility of the redefined object would be
   questionable.  Alternatively, ifNumber could be deprecated and not
   replaced.  However, the deprecation of ifNumber would require a
   change to that portion of ifIndex's definition which refers to
   ifNumber.  So, since the definition of ifIndex must be changed anyway
   in order to solve the problem, changes to ifNumber do not benefit the
   solution.

   The solution adopted in this memo is just to delete the requirement
   that the value of ifIndex must be less than the value of ifNumber,
   and to retain ifNumber with its current definition.  This is a minor
   change in the semantics of ifIndex; however, all existing agent
   implementations conform to this new definition, and in the interests
   of not requiring changes to existing agent implementations and to the
   many existing media-specific MIBs, this memo assumes that this change
   does not require ifIndex to be deprecated.  Experience indicates that
   this assumption does "break" a few management applications, but this
   is considered preferable to breaking all agent implementations.

   This solution also results in the possibility of "holes" in the
   ifTable, i.e., the ifIndex values of conceptual rows in the ifTable
   are not necessarily contiguous, but SNMP's GetNext (and GetBulk)
   operation easily deals with such holes.  The value of ifNumber still
   represents the number of conceptual rows, which increases/decreases
   as new interfaces are dynamically added/removed.

   The requirement for constancy (between re-initializations) of an
   interface's ifIndex value is met by requiring that after an interface
   is dynamically removed, its ifIndex value is not re-used by a
   *different* dynamically added interface until after the following
   re-initialization of the network management system.  This avoids the
   need for assignment (in advance) of ifIndex values for all possible
   interfaces that might be added dynamically.  The exact meaning of a
   "different" interface is hard to define, and there will be gray
   areas.  Any firm definition in this document would likely turn out to
   be inadequate.  Instead, implementors must choose what it means in
   their particular situation, subject to the following rules:
Top   ToC   RFC2863 - Page 12
   (1)   a previously-unused value of ifIndex must be assigned to a
         dynamically added interface if an agent has no knowledge of
         whether the interface is the "same" or "different" to a
         previously incarnated interface.

   (2)   a management station, not noticing that an interface has gone
         away and another has come into existence, must not be confused
         when calculating the difference between the counter values
         retrieved on successive polls for a particular ifIndex value.

   When the new interface is the same as an old interface, but a
   discontinuity in the value of the interface's counters cannot be
   avoided, the ifTable has (until now) required that a new ifIndex
   value be assigned to the returning interface.  That is, either all
   counter values have had to be retained during the absence of an
   interface in order to use the same ifIndex value on that interface's
   return, or else a new ifIndex value has had to be assigned to the
   returning interface.  Both alternatives have proved to be burdensome
   to some implementations:

   (1)   maintaining the counter values may not be possible (e.g., if
         they are maintained on removable hardware),

   (2)   using a new ifIndex value presents extra work for management
         applications.  While the potential need for such extra work is
         unavoidable on agent re-initializations, it is desirable to
         avoid it between re-initializations.

   To address this, a new object, ifCounterDiscontinuityTime, has been
   defined to record the time of the last discontinuity in an
   interface's counters.  By monitoring the value of this new object, a
   management application can now detect counter discontinuities without
   the ifIndex value of the interface being changed.  Thus, an agent
   which implements this new object should, when a new interface is the
   same as an old interface, retain that interface's ifIndex value and
   update if necessary the interface's value of
   ifCounterDiscontinuityTime.  With this new object, a management
   application must, when calculating differences between counter values
   retrieved on successive polls, discard any calculated difference for
   which the value of ifCounterDiscontinuityTime is different for the
   two polls.  (Note that this test must be performed in addition to the
   normal checking of sysUpTime to detect an agent re-initialization.)
   Since such discards are a waste of network management processing and
   bandwidth, an agent should not update the value of
   ifCounterDiscontinuityTime unless absolutely necessary.

   While defining this new object is a change in the semantics of the
   ifTable counter objects, it is impractical to deprecate and redefine
Top   ToC   RFC2863 - Page 13
   all these counters because of their wide deployment and importance.
   Also, a survey of implementations indicates that many agents and
   management applications do not correctly implement this aspect of the
   current semantics (because of the burdensome issues mentioned above),
   such that the practical implications of such a change is small.
   Thus, this breach of the SMI's rules is considered to be acceptable.

   Note, however, that the addition of ifCounterDiscontinuityTime does
   not change the fact that:

      it is necessary at certain times for the assignment of
      ifIndex values to change on a re-initialization of the agent
      (such as a reboot).

   The possibility of ifIndex value re-assignment must be accommodated
   by a management application whenever the value of sysUpTime is reset
   to zero.

   Note also that some agents support multiple "naming scopes", e.g.,
   for an SNMPv1 agent, multiple values of the SNMPv1 community string.
   For such an agent (e.g., a CNM agent which supports a different
   subset of interfaces for different customers), there is no required
   relationship between the ifIndex values which identify interfaces in
   one naming scope and those which identify interfaces in another
   naming scope.  It is the agent's choice as to whether the same or
   different ifIndex values identify the same or different interfaces in
   different naming scopes.

   Because of the restriction of the value of ifIndex to be less than
   ifNumber, interfaces have been numbered with small integer values.
   This has led to the ability by humans to use the ifIndex values as
   (somewhat) user-friendly names for network interfaces (e.g.,
   "interface number 3").  With the relaxation of the restriction on the
   value of ifIndex, there is now the possibility that ifIndex values
   could be assigned as very large numbers (e.g., memory addresses).
   Such numbers would be much less user-friendly.  Therefore, this memo
   recommends that ifIndex values still be assigned as (relatively)
   small integer values starting at 1, even though the values in use at
   any one time are not necessarily contiguous.  (Note that this makes
   remembering which values have been assigned easy for agents which
   dynamically add new interfaces)

   A new problem is introduced by representing each sub-layer as an
   ifTable entry.  Previously, there usually was a simple, direct,
   mapping of interfaces to the physical ports on systems.  This mapping
   would be based on the ifIndex value.  However, by having an ifTable
   entry for each interface sub-layer, mapping from interfaces to
   physical ports becomes increasingly problematic.
Top   ToC   RFC2863 - Page 14
   To address this issue, a new object, ifName, is added to the MIB.
   This object contains the device's local name (e.g., the name used at
   the device's local console) for the interface of which the relevant
   entry in the ifTable is a component.  For example, consider a router
   having an interface composed of PPP running over an RS-232 port.  If
   the router uses the name "wan1" for the (combined) interface, then
   the ifName objects for the corresponding PPP and RS-232 entries in
   the ifTable would both have the value "wan1".  On the other hand, if
   the router uses the name "wan1.1" for the PPP interface and "wan1.2"
   for the RS-232 port, then the ifName objects for the corresponding
   PPP and RS-232 entries in the ifTable would have the values "wan1.1"
   and "wan1.2", respectively.  As an another example, consider an agent
   which responds to SNMP queries concerning an interface on some other
   (proxied) device:  if such a proxied device associates a particular
   identifier with an interface, then it is appropriate to use this
   identifier as the value of the interface's ifName, since the local
   console in this case is that of the proxied device.

   In contrast, the existing ifDescr object is intended to contain a
   description of an interface, whereas another new object, ifAlias,
   provides a location in which a network management application can
   store a non-volatile interface-naming value of its own choice.  The
   ifAlias object allows a network manager to give one or more
   interfaces their own unique names, irrespective of any interface-
   stack relationship.  Further, the ifAlias name is non-volatile, and
   thus an interface must retain its assigned ifAlias value across
   reboots, even if an agent chooses a new ifIndex value for the
   interface.

3.1.6. Counter Size

As the speed of network media increase, the minimum time in which a 32 bit counter will wrap decreases. For example, a 10Mbs stream of back-to-back, full-size packets causes ifInOctets to wrap in just over 57 minutes; at 100Mbs, the minimum wrap time is 5.7 minutes, and at 1Gbs, the minimum is 34 seconds. Requiring that interfaces be polled frequently enough not to miss a counter wrap is increasingly problematic. A rejected solution to this problem was to scale the counters; for example, ifInOctets could be changed to count received octets in, say, 1024 byte blocks. While it would provide acceptable functionality at high rates of the counted-events, at low rates it suffers. If there is little traffic on an interface, there might be a significant interval before enough of the counted-events occur to cause the scaled counter to be incremented. Traffic would then appear to be very bursty, leading to incorrect conclusions of the network's performance.
Top   ToC   RFC2863 - Page 15
   Instead, this memo adopts expanded, 64 bit, counters.  These counters
   are provided in new "high capacity" groups.  The old, 32-bit,
   counters have not been deprecated.  The 64-bit counters are to be
   used only when the 32-bit counters do not provide enough capacity;
   that is, when the 32 bit counters could wrap too fast.

   For interfaces that operate at 20,000,000 (20 million) bits per
   second or less, 32-bit byte and packet counters MUST be supported.
   For interfaces that operate faster than 20,000,000 bits/second, and
   slower than 650,000,000 bits/second, 32-bit packet counters MUST be
   supported and 64-bit octet counters MUST be supported.  For
   interfaces that operate at 650,000,000 bits/second or faster, 64-bit
   packet counters AND 64-bit octet counters MUST be supported.

   These speed thresholds were chosen as reasonable compromises based on
   the following:

   (1)   The cost of maintaining 64-bit counters is relatively high, so
         minimizing the number of agents which must support them is
         desirable.  Common interfaces (such as 10Mbs Ethernet) should
         not require them.

   (2)   64-bit counters are a new feature, introduced in the SMIv2.  It
         is reasonable to expect that support for them will be spotty
         for the immediate future.  Thus, we wish to limit them to as
         few systems as possible.  This, in effect, means that 64-bit
         counters should be limited to higher speed interfaces.
         Ethernet (10,000,000 bps) and Token Ring (16,000,000 bps) are
         fairly wide-spread so it seems reasonable to not require 64-bit
         counters for these interfaces.

   (3)   The 32-bit octet counters will wrap in the following times, for
         the following interfaces (when transmitting maximum-sized
         packets back-to-back):

         -   10Mbs Ethernet: 57 minutes,

         -   16Mbs Token Ring: 36 minutes,

         -   a US T3 line (45 megabits): 12 minutes,

         -   FDDI: 5.7 minutes

   (4)   The 32-bit packet counters wrap in about 57 minutes when 64-
         byte packets are transmitted back-to-back on a 650,000,000
         bit/second link.
Top   ToC   RFC2863 - Page 16
   As an aside, a 1-terabit/second (1,000 Gbs) link will cause a 64 bit
   octet counter to wrap in just under 5 years.  Conversely, an
   81,000,000 terabit/second link is required to cause a 64-bit counter
   to wrap in 30 minutes.  We believe that, while technology rapidly
   marches forward, this link speed will not be achieved for at least
   several years, leaving sufficient time to evaluate the introduction
   of 96 bit counters.

   When 64-bit counters are in use, the 32-bit counters MUST still be
   available.  They will report the low 32-bits of the associated 64-bit
   count (e.g., ifInOctets will report the least significant 32 bits of
   ifHCInOctets).  This enhances inter-operability with existing
   implementations at a very minimal cost to agents.

   The new "high capacity" groups are:

   (1)   the ifHCFixedLengthGroup for character-oriented/fixed-length
         interfaces, and the ifHCPacketGroup for packet-based
         interfaces; both of these groups include 64 bit counters for
         octets, and

   (2)   the ifVHCPacketGroup for packet-based interfaces; this group
         includes 64 bit counters for octets and packets.

3.1.7. Interface Speed

Network speeds are increasing. The range of ifSpeed is limited to reporting a maximum speed of (2**31)-1 bits/second, or approximately 2.2Gbs. SONET defines an OC-48 interface, which is defined at operating at 48 times 51 Mbs, which is a speed in excess of 2.4Gbs. Thus, ifSpeed is insufficient for the future, and this memo defines an additional object: ifHighSpeed. The ifHighSpeed object reports the speed of the interface in 1,000,000 (1 million) bits/second units. Thus, the true speed of the interface will be the value reported by this object, plus or minus 500,000 bits/second. Other alternatives considered (but rejected) were: (1) Making the interface speed a 64-bit gauge. This was rejected since the current SMI does not allow such a syntax. Furthermore, even if 64-bit gauges were available, their use would require additional complexity in agents due to an increased requirement for 64-bit operations.
Top   ToC   RFC2863 - Page 17
   (2)   We also considered making "high-32 bit" and "low-32-bit"
         objects which, when combined, would be a 64-bit value.  This
         simply seemed overly complex for what we are trying to do.

      Furthermore, a full 64-bits of precision does not seem necessary.
      The value of ifHighSpeed will be the only report of interface
      speed for interfaces that are faster than 4,294,967,295 bits per
      second.  At this speed, the granularity of ifHighSpeed will be
      1,000,000 bits per second, thus the error will be 1/4294, or about
      0.02%.  This seems reasonable.

   (3)   Adding a "scale" object, which would define the units which
         ifSpeed's value is.

      This would require two additional objects; one for the scaling
      object, and one to replace the current ifSpeed.  This later object
      is required since the semantics of ifSpeed would be significantly
      altered, and manager stations which do not understand the new
      semantics would be confused.

3.1.8. Multicast/Broadcast Counters

In MIB-II, the ifTable counters for multicast and broadcast packets are combined as counters of non-unicast packets. In contrast, the ifExtensions MIB [19] defined one set of counters for multicast, and a separate set for broadcast packets. With the separate counters, the original combined counters become redundant. To avoid this redundancy, the non-unicast counters are deprecated. For the output broadcast and multicast counters defined in RFC 1229, their definitions varied slightly from the packet counters in the ifTable, in that they did not count errors/discarded packets. Thus, this memo defines new objects with better aligned definitions. Counters with 64 bits of range are also needed, as explained above.

3.1.9. Trap Enable

In the multi-layer interface model, each sub-layer for which there is an entry in the ifTable can generate linkUp/linkDown Traps. Since interface state changes would tend to propagate through the interface (from top to bottom, or bottom to top), it is likely that several traps would be generated for each linkUp/linkDown occurrence. It is desirable to provide a mechanism for manager stations to control the generation of these traps. To this end, the ifLinkUpDownTrapEnable object has been added. This object allows managers to limit generation of traps to just the sub-layers of interest.
Top   ToC   RFC2863 - Page 18
   The default setting should limit the number of traps generated to one
   per interface per linkUp/linkDown event.  Furthermore, it seems that
   the state changes of most interest to network managers occur at the
   lowest level of an interface stack.  Therefore we specify that by
   default, only the lowest sub-layer of the interface generate traps.

3.1.10. Addition of New ifType values

Over time, there is the need to add new ifType enumerated values for new interface types. If the syntax of ifType were defined in the MIB in section 6, then a new version of this MIB would have to be re- issued in order to define new values. In the past, re-issuing of a MIB has occurred only after several years. Therefore, the syntax of ifType is changed to be a textual convention, such that the enumerated integer values are now defined in the textual convention, IANAifType, defined in a different document. This allows additional values to be documented without having to re-issue a new version of this document. The Internet Assigned Number Authority (IANA) is responsible for the assignment of all Internet numbers, including various SNMP-related numbers, and specifically, new ifType values.

3.1.11. InterfaceIndex Textual Convention

A new textual convention, InterfaceIndex, has been defined. This textual convention "contains" all of the semantics of the ifIndex object. This allows other MIB modules to easily import the semantics of ifIndex.

3.1.12. New states for IfOperStatus

Three new states have been added to ifOperStatus: 'dormant', 'notPresent', and 'lowerLayerDown'. The dormant state indicates that the relevant interface is not actually in a condition to pass packets (i.e., it is not 'up') but is in a "pending" state, waiting for some external event. For "on- demand" interfaces, this new state identifies the situation where the interface is waiting for events to place it in the up state. Examples of such events might be: (1) having packets to transmit before establishing a connection to a remote system; (2) having a remote system establish a connection to the interface (e.g. dialing up to a slip-server).
Top   ToC   RFC2863 - Page 19
   The notPresent state is a refinement on the down state which
   indicates that the relevant interface is down specifically because
   some component (typically, a hardware component) is not present in
   the managed system.  Examples of use of the notPresent state are:

   (1)   to allow an interface's conceptual row including its counter
         values to be retained across a "hot swap" of a card/module,
         and/or

   (2)   to allow an interface's conceptual row to be created, and
         thereby enable interfaces to be pre-configured prior to
         installation of the hardware needed to make the interface
         operational.

   Agents are not required to support interfaces in the notPresent
   state.  However, from a conceptual viewpoint, when a row in the
   ifTable is created, it first enters the notPresent state and then
   subsequently transitions into the down state; similarly, when a row
   in the ifTable is deleted, it first enters the notPresent state and
   then subsequently the object instances are deleted.  For an agent
   with no support for notPresent, both of these transitions (from the
   notPresent state to the down state, and from the notPresent state to
   the instances being removed) are immediate, i.e., the transition does
   not last long enough to be recorded by ifOperStatus.  Even for those
   agents which do support interfaces in the notPresent state, the
   length of time and conditions under which an interface stays in the
   notPresent state is implementation-specific.

   The lowerLayerDown state is also a refinement on the down state.
   This new state indicates that this interface runs "on top of" one or
   more other interfaces (see ifStackTable) and that this interface is
   down specifically because one or more of these lower-layer interfaces
   are down.

3.1.13. IfAdminStatus and IfOperStatus

The down state of ifOperStatus now has two meanings, depending on the value of ifAdminStatus. (1) if ifAdminStatus is not down and ifOperStatus is down then a fault condition is presumed to exist on the interface. (2) if ifAdminStatus is down, then ifOperStatus will normally also be down (or notPresent) i.e., there is not (necessarily) a fault condition on the interface. Note that when ifAdminStatus transitions to down, ifOperStatus will normally also transition to down. In this situation, it is possible
Top   ToC   RFC2863 - Page 20
   that ifOperStatus's transition will not occur immediately, but rather
   after a small time lag to complete certain operations before going
   "down"; for example, it might need to finish transmitting a packet.
   If a manager station finds that ifAdminStatus is down and
   ifOperStatus is not down for a particular interface, the manager
   station should wait a short while and check again.  If the condition
   still exists, only then should it raise an error indication.
   Naturally, it should also ensure that ifLastChange has not changed
   during this interval.

   Whenever an interface table entry is created (usually as a result of
   system initialization), the relevant instance of ifAdminStatus is set
   to down, and ifOperStatus will be down or notPresent.

   An interface may be enabled in two ways: either as a result of
   explicit management action (e.g. setting ifAdminStatus to up) or as a
   result of the managed system's initialization process.  When
   ifAdminStatus changes to the up state, the related ifOperStatus
   should do one of the following:

   (1)   Change to the up state if and only if the interface is able to
         send and receive packets.

   (2)   Change to the lowerLayerDown state if and only if the interface
         is prevented from entering the up state because of the state of
         one or more of the interfaces beneath it in the interface
         stack.

   (3)   Change to the dormant state if and only if the interface is
         found to be operable, but the interface is waiting for other,
         external, events to occur before it can transmit or receive
         packets.  Presumably when the expected events occur, the
         interface will then change to the up state.

   (4)   Remain in the down state if an error or other fault condition
         is detected on the interface.

   (5)   Change to the unknown state if, for some reason, the state of
         the interface can not be ascertained.

   (6)   Change to the testing state if some test(s) must be performed
         on the interface. Presumably after completion of the test, the
         interface's state will change to up, dormant, or down, as
         appropriate.

   (7)   Remain in the notPresent state if interface components are
         missing.
Top   ToC   RFC2863 - Page 21

3.1.14. IfOperStatus in an Interface Stack

When an interface is a part of an interface-stack, but is not the lowest interface in the stack, then: (1) ifOperStatus has the value 'up' if it is able to pass packets due to one or more interfaces below it in the stack being 'up', irrespective of whether other interfaces below it are 'down', ' dormant', 'notPresent', 'lowerLayerDown', 'unknown' or ' testing'. (2) ifOperStatus may have the value 'up' or 'dormant' if one or more interfaces below it in the stack are 'dormant', and all others below it are either 'down', 'dormant', 'notPresent', ' lowerLayerDown', 'unknown' or 'testing'. (3) ifOperStatus has the value 'lowerLayerDown' while all interfaces below it in the stack are either 'down', ' notPresent', 'lowerLayerDown', or 'testing'.

3.1.15. Traps

The exact definition of when linkUp and linkDown traps are generated has been changed to reflect the changes to ifAdminStatus and ifOperStatus. Operational experience indicates that management stations are most concerned with an interface being in the down state and the fact that this state may indicate a failure. Thus, it is most useful to instrument transitions into/out of either the up state or the down state. Instrumenting transitions into or out of the up state was rejected since it would have the drawback that a demand interface might have many transitions between up and dormant, leading to many linkUp traps and no linkDown traps. Furthermore, if a node's only interface is the demand interface, then a transition to dormant would entail generation of a linkDown trap, necessitating bringing the link to the up state (and a linkUp trap)!! On the other hand, instrumenting transitions into or out of the down state (to/from all other states except notPresent) has the advantages: (1) A transition into the down state (from a state other than notPresent) will occur when an error is detected on an interface. Error conditions are presumably of great interest to network managers.
Top   ToC   RFC2863 - Page 22
   (2)   Departing the down state (to a state other than the notPresent
         state) generally indicates that the interface is going to
         either up or dormant, both of which are considered "healthy"
         states.

   Furthermore, it is believed that generating traps on transitions into
   or out of the down state (except to/from the notPresent state) is
   generally consistent with current usage and interpretation of these
   traps by manager stations.

   Transitions to/from the notPresent state are concerned with the
   insertion and removal of hardware, and are outside the scope of these
   traps.

   Therefore, this memo defines that LinkUp and linkDown traps are
   generated just after ifOperStatus leaves, or just before it enters,
   the down state, respectively; except that LinkUp and linkDown traps
   are never generated on transitions to/from the notPresent state.  For
   the purpose of deciding when these traps occur, the lowerLayerDown
   state and the down state are considered to be equivalent, i.e., there
   is no trap on transition from lowerLayerDown into down, and there is
   a trap on transition from any other state except down (and
   notPresent) into lowerLayerDown.

   Note that this definition allows a node with only one interface to
   transmit a linkDown trap before that interface goes down.  (Of
   course, when the interface is going down because of a failure
   condition, the linkDown trap probably cannot be successfully
   transmitted anyway.)

   Some interfaces perform a link "training" function when trying to
   bring the interface up.  In the event that such an interface were
   defective, then the training function would fail and the interface
   would remain down, and the training function might be repeated at
   appropriate intervals.  If the interface, while performing this
   training function, were considered to the in the testing state, then
   linkUp and linkDown traps would be generated for each start and end
   of the training function.  This is not the intent of the linkUp and
   linkDown traps, and therefore, while performing such a training
   function, the interface's state should be represented as down.

   An exception to the above generation of linkUp/linkDown traps on
   changes in ifOperStatus, occurs when an interface is "flapping",
   i.e., when it is rapidly oscillating between the up and down states.
   If traps were generated for each such oscillation, the network and
   the network management system would be flooded with unnecessary
   traps.  In such a situation, the agent should limit the rate at which
   it generates traps.
Top   ToC   RFC2863 - Page 23

3.1.16. ifSpecific

The original definition of the OBJECT IDENTIFIER value of ifSpecific was not sufficiently clear. As a result, different implementors used it differently, and confusion resulted. Some implementations set the value of ifSpecific to the OBJECT IDENTIFIER that defines the media- specific MIB, i.e., the "foo" of: foo OBJECT IDENTIFIER ::= { transmission xxx } while others set it to be OBJECT IDENTIFIER of the specific table or entry in the appropriate media-specific MIB (i.e., fooTable or fooEntry), while still others set it be the OBJECT IDENTIFIER of the index object of the table's row, including instance identifier, (i.e., fooIfIndex.ifIndex). A definition based on the latter would not be sufficient unless it also allowed for media-specific MIBs which include several tables, where each table has its own (different) indexing. The only definition that can both be made explicit and can cover all the useful situations is to have ifSpecific be the most general value for the media-specific MIB module (the first example given above). This effectively makes it redundant because it contains no more information than is provided by ifType. Thus, ifSpecific has been deprecated.

3.1.17. Creation/Deletion of Interfaces

While some interfaces, for example, most physical interfaces, cannot be created via network management, other interfaces such as logical interfaces sometimes can be. The ifTable contains only generic information about an interface. Almost all 'create-able' interfaces have other, media-specific, information through which configuration parameters may be supplied prior to creating such an interface. Thus, the ifTable does not itself support the creation or deletion of an interface (specifically, it has no RowStatus [6] column). Rather, if a particular interface type supports the dynamic creation and/or deletion of an interface of that type, then that media-specific MIB should include an appropriate RowStatus object (see the ATM LAN- Emulation Client MIB [20] for an example of a MIB which does this). Typically, when such a RowStatus object is created/deleted, then the conceptual row in the ifTable appears/disappears as a by-product, and an ifIndex value (chosen by the agent) is stored in an appropriate object in the media-specific MIB.
Top   ToC   RFC2863 - Page 24

3.1.18. All Values Must be Known

There are a number of situations where an agent does not know the value of one or more objects for a particular interface. In all such circumstances, an agent MUST NOT instantiate an object with an incorrect value; rather, it MUST respond with the appropriate error/exception condition (e.g., noSuchInstance or noSuchName). One example is where an agent is unable to count the occurrences defined by one (or more) of the ifTable counters. In this circumstance, the agent MUST NOT instantiate the particular counter with a value of, say, zero. To do so would be to provide mis- information to a network management application reading the zero value, and thereby assuming that there have been no occurrences of the event (e.g., no input errors because ifInErrors is always zero). Sometimes the lack of knowledge of an object's value is temporary. For example, when the MTU of an interface is a configured value and a device dynamically learns the configured value through (after) exchanging messages over the interface (e.g., ATM LAN-Emulation [20]). In such a case, the value is not known until after the ifTable entry has already been created. In such a case, the ifTable entry should be created without an instance of the object whose value is unknown; later, when the value becomes known, the missing object can then be instantiated (e.g., the instance of ifMtu is only instantiated once the interface's MTU becomes known). As a result of this "known values" rule, management applications MUST be able to cope with the responses to retrieving the object instances within a conceptual row of the ifTable revealing that some of the row's columnar objects are missing/not available.

4. Media-Specific MIB Applicability

The exact use and semantics of many objects in this MIB are open to some interpretation. This is a result of the generic nature of this MIB. It is not always possible to come up with specific, unambiguous, text that covers all cases and yet preserves the generic nature of the MIB. Therefore, it is incumbent upon a media-specific MIB designer to, wherever necessary, clarify the use of the objects in this MIB with respect to the media-specific MIB.
Top   ToC   RFC2863 - Page 25
   Specific areas of clarification include

   Layering Model
      The media-specific MIB designer MUST completely and unambiguously
      specify the layering model used.  Each individual sub-layer must
      be identified, as must the ifStackTable's portrayal of the
      relationship(s) between the sub-layers.

   Virtual Circuits
      The media-specific MIB designer MUST specify whether virtual
      circuits are assigned entries in the ifTable or not.  If they are,
      compelling rationale must be presented.

   ifRcvAddressTable
      The media-specific MIB designer MUST specify the applicability of
      the ifRcvAddressTable.

   ifType
      For each of the ifType values to which the media-specific MIB
      applies, it must specify the mapping of ifType values to media-
      specific MIB module(s) and instances of MIB objects within those
      modules.

   ifXxxOctets
      The definitions of ifInOctets and ifOutOctets (and similarly,
      ifHCInOctets and ifHCOutOctets) specify that their values include
      framing characters.  The media-specific MIB designer MUST specify
      any special conditions of the media concerning the inclusion of
      framing characters, especially with respect to frames with errors.

   However, wherever this interface MIB is specific in the semantics,
   DESCRIPTION, or applicability of objects, the media-specific MIB
   designer MUST NOT change said semantics, DESCRIPTION, or
   applicability.

5. Overview

This MIB consists of 4 tables: ifTable This table is the ifTable from MIB-II. ifXTable This table contains objects that have been added to the Interface MIB as a result of the Interface Evolution effort, or replacements for objects of the original (MIB-II) ifTable that were deprecated
Top   ToC   RFC2863 - Page 26
      because the semantics of said objects have significantly changed.
      This table also contains objects that were previously in the
      ifExtnsTable.

   ifStackTable
      This table contains objects that define the relationships among
      the sub-layers of an interface.

   ifRcvAddressTable
      This table contains objects that are used to define the media-
      level addresses which this interface will receive.  This table is
      a generic table.  The designers of media-specific MIBs must define
      exactly how this table applies to their specific MIB.



(page 26 continued on part 2)

Next Section