4.  SMIv2 Usage Guidelines

   In general, MIB modules in IETF standards-track specifications MUST
   comply with all syntactic and semantic requirements of SMIv2
   [RFC2578] [RFC2579] [RFC2580] that apply to "standard" MIB modules
   and except as noted below SHOULD comply with SMIv2 recommendations.
   The guidelines in this section are intended to supplement the SMIv2
   documents in the following ways:

    o to document the current generally accepted interpretation when
      those documents contain ambiguities or contradictions;

    o to update recommendations in those documents that have been shown
      by practical experience to be out-of-date or otherwise suboptimal;

    o to provide guidance in selection of SMIv2 options in cases where
      there is a consensus on a preferred approach.

4.1.  Module Names

   RFC 2578 Section 3 specifies the rules for module names.  Note in
   particular that names of "standard" modules MUST be unique, MUST
   follow the syntax rules in RFC 2578 Section 3, and MUST NOT be
   changed when a MIB module is revised (see also RFC 2578 Section 10).

   It is RECOMMENDED that module names be mnemonic.  See Appendix C for
   suggested naming conventions.

4.2.  Descriptors, TC Names, and Labels

   RFC 2578 Sections 3.1, 7.1.1, and 7.1.4 and RFC 2579 Section 3
   recommend that descriptors and names associated with macro
   invocations and labels associated with enumerated INTEGER and BITS
   values be no longer than 32 characters, but require that they be no
   longer than 64 characters.

   Restricting descriptors, TC names, and labels to 32 characters often
   conflicts with the recommendation that they be mnemonic and (for
   descriptors and TC names) with the requirement that they be unique
   (see RFC 2578 Section 3.1 and RFC 2579 Section 3).  The consensus of
   the current pool of MIB reviewers is that the SMIv2 recommendation to
   limit descriptors, TC names, and labels to 32 characters SHOULD be
   set aside in favor of promoting clarity and uniqueness and that
   automated tools such as MIB compilers SHOULD NOT by default generate
   warnings for violating that recommendation.

   Note that violations of the 64-character limit MUST NOT be ignored;
   they MUST be treated as errors.

   See Appendix C for suggested descriptor and TC naming conventions.

4.3.  Naming Hierarchy

   RFC 2578 Section 4 describes the object identifier subtrees that are
   maintained by IANA and specifies the usages for those subtrees.  In
   particular, the mgmt subtree { iso 3 6 1 2 } is used to identify IETF
   "standard" objects, while the experimental subtree { iso 3 6 1 3 } is
   used to identify objects that are under development in the IETF.  It
   is REQUIRED that objects be moved from the experimental subtree to
   the mgmt subtree when a MIB module enters the IETF standards track.

   Experience has shown that it is impractical to move objects from one
   subtree to another once those objects have seen large-scale use in an
   operational environment.  Hence any object that is targeted for
   deployment in an operational environment MUST NOT be registered under
   the experimental subtree, irrespective of the standardization status
   of that object.  The experimental subtree should be used only for
   objects that are intended for limited experimental deployment.  Such
   objects typically are defined in Experimental RFCs.

   Note:  the term "object", as used here and in RFC 2578 Section 4, is
   to be broadly interpreted as any construct that results in an OBJECT
   IDENTIFIER registration.  The list of such constructs is specified in
   RFC 2578 Section 3.6.

4.4.  IMPORTS Statement

   RFC 2578 Section 3.2 specifies which symbols must be imported and
   also lists certain predefined symbols that must not be imported.

   The general requirement is that if an external symbol other than a
   predefined ASN.1 type or the BITS construct is used, then it MUST be
   mentioned in the module's IMPORTS statement.  The words "external
   object" in the first paragraph of that section may give the

   impression that such symbols are limited to those that refer to
   object definitions, but that is not the case, as subsequent
   paragraphs should make clear.

   Note that exemptions to this general requirement are granted by RFC
   2580 Sections 5.4.3 and 6.5.2 for descriptors of objects appearing in
   the OBJECT clause of a MODULE-COMPLIANCE statement or in the
   VARIATION clause of an AGENT-CAPABILITIES statement.  Some MIB
   compilers also grant exemptions to descriptors of notifications
   appearing in a VARIATION clause and to descriptors of object groups
   and notification groups referenced by a MANDATORY-GROUPS clause, a
   GROUP clause, or an INCLUDES clause, although RFC 2580 (through
   apparent oversight) does not mention those cases.  The exemptions are
   sometimes seen as unhelpful because they make IMPORTS rules more
   complicated and inter-module dependencies less obvious than they
   otherwise would be.  External symbols referenced by compliance
   statements and capabilities statements MAY therefore be listed in the
   IMPORTS statement; if this is done, it SHOULD be done consistently.

   Finally, even though it is not forbidden by the SMI, it is considered
   poor style to import symbols that are not used, and standards-track
   MIB modules SHOULD NOT do so.

4.5.  MODULE-IDENTITY Invocation

   RFC 2578 Section 3 requires that all SMIv2 MIB modules start with
   exactly one invocation of the MODULE-IDENTITY macro.  This invocation
   MUST appear immediately after the IMPORTS statement.

   RFC 2578 Section 5 describes how the various clauses are used.  The
   following additional guidelines apply to all MIB modules over which
   the IETF has change control:

   - If the module was developed by an IETF working group, then the
     ORGANIZATION clause MUST provide the full name of the working
     group, and the CONTACT-INFO clause MUST include working group
     mailing list information.  The CONTACT-INFO clause SHOULD also
     provide a pointer to the working group's web page.

   - A REVISION clause MUST be present for each revision of the MIB
     module, and the UTC time of the most recent REVISION clause MUST
     match that of the LAST-UPDATED clause.  The DESCRIPTION clause
     associated with each revision MUST state in which RFC that revision
     appeared and SHOULD provide a list of all significant changes.
     When a MIB module is revised, UTC times in all REVISION clauses
     SHOULD be updated to use four-digit year notation.

   - The value assigned to the MODULE-IDENTITY descriptor MUST be unique
     and (for IETF standards-track MIB modules) SHOULD reside under the
     mgmt subtree [RFC2578].  Most often it will be an IANA-assigned
     value directly under mib-2 [RFC2578], although for media-specific
     MIB modules that extend the IF-MIB [RFC2863] it is customary to use
     an IANA-assigned value under transmission [RFC2578].  In the past,
     some IETF working groups have made their own assignments from
     subtrees delegated to them by IANA, but that practice has proven
     problematic and is NOT RECOMMENDED.

   While a MIB module is under development, the RFC number in which it
   will eventually be published is usually unknown and must be filled in
   by the RFC Editor prior to publication.  An appropriate form for the
   REVISION clause applying to a version under development would be
   something along the following lines:

          REVISION    "200212132358Z"  -- December 13, 2002
          DESCRIPTION "Initial version, published as RFC yyyy."
   -- RFC Ed.: replace yyyy with actual RFC number & remove this note

   Note that after RFC publication, a REVISION clause is present only
   for published versions of a MIB module and not for interim versions
   that existed only as Internet-Drafts.  Thus, a draft version of a MIB
   module MUST contain just one new REVISION clause that covers all
   changes since the last published version (if any).

   When the initial version of a MIB module is under development, the
   value assigned to the MODULE-IDENTITY descriptor will be unknown if
   an IANA-assigned value is used, because the assignment is made just
   prior to publication as an RFC.  The accepted form for the MODULE-
   IDENTITY statement in draft versions of such a module is something
   along the following lines:

      <descriptor> MODULE-IDENTITY

          [ ... ]

          ::= { <subtree> XXX }
   -- RFC Ed.: replace XXX with IANA-assigned number & remove this note

   where <descriptor> is whatever descriptor has been selected for the
   module and <subtree> is the subtree under which the module is to be
   registered (e.g., mib-2 or transmission).  Note that XXX must be
   temporarily replaced by a number in order for the module to compile.

   Note well:  prior to official assignment by the IANA, a draft
   document MUST use a placeholder (such as "XXX" above) rather than an
   actual number.  If trial implementations are desired during the
   development process, then an assignment under the 'experimental'
   subtree may be obtained from the IANA (cf. Section 4.3).

4.6.  Textual Conventions and Object Definitions

4.6.1.  Usage of Data Types

4.6.1.1.  INTEGER, Integer32, Gauge32, and Unsigned32

   The 32-bit integer data types INTEGER, Integer32, Gauge32, and
   Unsigned32 are described in RFC 2578 Section 2 and further elaborated
   in RFC 2578 Sections 7.1.1, 7.1.7, and 7.1.11.  The following
   guidelines apply when selecting one of these data types for an object
   definition or a textual convention:

   - For integer-valued enumerations:

     - INTEGER is REQUIRED; - Integer32, Unsigned32, and Gauge32 MUST
     NOT be used.

   Note that RFC 2578 recommends (but does not require) that integer-
   valued enumerations start at 1 and be numbered contiguously.  This
   recommendation SHOULD be followed unless there is a valid reason to
   do otherwise, e.g., to match values of external data or to indicate
   special cases, and any such special-case usage SHOULD be clearly
   documented.  For an example, see the InetAddressType TC [RFC4001].

   Although the SMI allows DEFVAL clauses for integer-valued
   enumerations to specify the default value either by label or by
   numeric value, the label form is preferred since all the examples in
   RFC 2578 are of that form and some tools do not accept the numeric
   form.

   - If the value range is between -2147483648..2147483647 (inclusive)
     and negative values are possible, then:

     - Integer32 is RECOMMENDED;
     - INTEGER is acceptable;
     - Unsigned32 and Gauge32 MUST NOT be used.

   - If the value range is between 0..4294967295 (inclusive) and the
     value of the information being modelled may increase above the
     maximum value or decrease below the minimum value, then:

     - Gauge32 is RECOMMENDED;
     - Unsigned32 is acceptable;
     - INTEGER and Integer32 MUST NOT be used if
       values greater than 2147483647 are possible.

   - If the value range is between 0..4294967295 (inclusive), and values
     greater than 2147483647 are possible, and the value of the
     information being modelled does not increase above the maximum
     value nor decrease below the minimum value, then:

     - Unsigned32 is RECOMMENDED;
     - Gauge32 is acceptable;
     - INTEGER and Integer32 MUST NOT be used.

   - If the value range is between 0..2147483647 (inclusive), and the
     value of the information being modelled does not increase above the
     maximum value nor decrease below the minimum value, then:

     - Unsigned32 is RECOMMENDED;
     - INTEGER, Integer32, and Gauge32 are acceptable.

   - For integer-valued objects that appear in an INDEX clause or for
     integer-valued TCs that are to be used in an index column:

     - Unsigned32 with a range that excludes zero is RECOMMENDED for
       most index objects.  It is acceptable to include zero in the
       range when it is semantically significant or when it is used as
       the index value for a unique row with special properties.  Such
       usage SHOULD be clearly documented in the DESCRIPTION clause.

     - Integer32 or INTEGER with a non-negative range is acceptable.
       Again, zero SHOULD be excluded from the range except when it is
       semantically significant or when it is used as the index value
       for a unique row with special properties, and in such cases the
       usage SHOULD be clearly documented in the DESCRIPTION clause.

     - Use of Gauge32 is acceptable for index objects that have gauge
       semantics.

   The guidelines above combine both the usage rules for integer data
   types and the INDEX rules in RFC 2578 Section 7.7 up to and including
   bullet (1) plus the next-to-last paragraph on page 28.

   Sometimes it will be necessary for external variables to represent
   values of an index object -- e.g., ifIndex [RFC2863].  In such cases,
   authors of the module containing that object SHOULD consider defining
   TCs such as InterfaceIndex and/or InterfaceIndexOrZero [RFC2863].

   Note that INTEGER is a predefined ASN.1 type and MUST NOT be present
   in a module's IMPORTS statement, whereas Integer32, Gauge32, and
   Unsigned32 are defined by SNMPv2-SMI and MUST be imported from that
   module if used.

4.6.1.2.  Counter32 and Counter64

   Counter32 and Counter64 have special semantics as described in RFC
   2578 Sections 7.1.6 and 7.1.10, respectively.  Object definitions
   MUST (and textual conventions SHOULD) respect these semantics.  That
   means:

   - It is OK to use Counter32/64 for counters that may/will be reset
     when the management subsystem is re-initialized or when other
     unusual/irregular events occur (e.g., counters maintained on a line
     card may be reset when the line card is reset).  However, if it is
     possible for such other unusual/irregular events to occur, the
     DESCRIPTION clause MUST state that this is so and MUST describe
     those other unusual/irregular events in sufficient detail that it
     is possible for a management application to determine whether a
     reset has occurred since the last time the counter was polled.  The
     RECOMMENDED way to do this is to provide a discontinuity indicator
     as described in RFC 2578 Sections 7.1.6 and 7.1.10.  For an example
     of such a discontinuity indicator, see the
     ifCounterDiscontinuityTime object in the IF-MIB [RFC2863].

   - It is NOT OK to put in the DESCRIPTION clause of a Counter32/64
     that there is a requirement that on a discontinuity the counter
     MUST reset to zero or to any other specific value.

   - It is NOT OK to put in the DESCRIPTION clause of a Counter32/64
     that there is a requirement that it MUST reset at any specific
     time/event (e.g., midnight).

   - It is NOT OK for one manager to request the agent to reset the
     value(s) of counter(s) to zero, and Counter32/64 is the wrong
     syntax for "counters" that regularly reset themselves to zero.  For
     the latter, it is better to define or use textual conventions such
     as those in RFC 3593 [RFC3593] or RFC 3705 [RFC3705].

   RFC 2578 Section 7.1.10 places a requirement on "standard" MIB
   modules that the Counter64 type may be used only if the information
   being modelled would wrap in less than one hour if the Counter32 type
   was used instead.  Now that SNMPv3 is an Internet Standard and SNMPv1
   is Historic (see http://www.rfc-editor.org/rfcxx00.html for status
   and [RFC3410] for rationale), there is no reason to continue
   enforcing this restriction.  Henceforth "standard" MIB modules MAY
   use the Counter64 type when it makes sense to do so, and MUST use

   Counter64 if the information being modelled would wrap in less than
   one hour if the Counter32 type was used instead.  Note also that
   there is no longer a requirement to define Counter32 counterparts for
   each Counter64 object, although one is still allowed to do so.

   There also exist closely-related textual conventions
   ZeroBasedCounter32 and ZeroBasedCounter64 defined in RMON2-MIB
   [RFC2021] and HCNUM-TC [RFC2856], respectively.

   The only difference between ZeroBasedCounter32/64 TCs and
   Counter32/64 is their starting value; at time=X, where X is their
   minimum-wrap-time after they were created, the behavior of
   ZeroBasedCounter32/64 becomes exactly the same as Counter32/64.
   Thus, the preceding paragraphs/rules apply not only to Counter32/64,
   but also to ZeroBasedCounter32/64 TCs.

4.6.1.3.  CounterBasedGauge64

   SMIv2 unfortunately does not provide 64-bit integer base types.  In
   order to make up for this omission, the CounterBasedGauge64 textual
   convention is defined in HCNUM-TC [RFC2856].  This TC uses Counter64
   as a base type, but discards the special counter semantics, which is
   allowed under the generally accepted interpretation of RFC 2579
   Section 3.3.  It does inherit all the syntactic restrictions of that
   type, which means that it MUST NOT be subtyped and that objects
   defined with it MUST NOT appear in an INDEX clause, MUST NOT have a
   DEFVAL clause, and MUST have a MAX-ACCESS of read-only or
   accessible-for-notify.

   This TC SHOULD be used for object definitions that require a 64-bit
   unsigned data type with gauge semantics.  If a 64-bit unsigned data
   type with different semantics is needed, then a different TC based on
   Counter64 MUST be used, since one TC cannot refine another (cf. RFC
   2579 Section 3.5).

4.6.1.4.  OCTET STRING

   The OCTET STRING type is described in RFC 2578 Section 7.1.2.  It
   represents arbitrary binary or textual data whose length is between 0
   and 65535 octets inclusive.  Objects and TCs whose SYNTAX is of this
   type SHOULD have a size constraint when the actual bounds are more
   restrictive than the SMI-imposed limits.  This is particularly true
   for index objects.  Note, however, that size constraints SHOULD NOT
   be imposed arbitrarily, as the SMI does not permit them to be changed
   afterward.

   There exist a number of standard TCs that cater to some of the more
   common requirements for specialized OCTET STRING types.  In
   particular, SNMPv2-TC [RFC2579] contains the DisplayString,
   PhysAddress, MacAddress, and DateAndTime TCs; the SNMP-FRAMEWORK-MIB
   [RFC3411] contains the SnmpAdminString TC; and the SYSAPPL-MIB
   [RFC2287] contains the Utf8String and LongUtf8String TCs.  When a
   standard TC provides the desired semantics, it SHOULD be used in an
   object's SYNTAX clause instead of OCTET STRING or an equivalent
   locally-defined TC.

   Note that OCTET STRING is a predefined ASN.1 type and MUST NOT be
   present in a module's IMPORTS statement.

4.6.1.5.  OBJECT IDENTIFIER

   The OBJECT IDENTIFIER type is described in RFC 2578 Section 7.1.3.
   Its instances represent administratively assigned names.  Note that
   both the SMI and the SNMP protocol limit instances of this type to
   128 sub-identifiers and require that each sub-identifier be within
   the range 0 to 4294967295 inclusive.  Subtyping is not allowed.

   The purpose of OBJECT IDENTIFIER values is to provide authoritative
   identification either for some type of item or for a specific
   instance of some type of item.  Among the items that can be
   identified in this way are definitions in MIB modules created via the
   MODULE-IDENTITY, OBJECT-IDENTITY, OBJECT-TYPE, NOTIFICATION-TYPE,
   OBJECT-GROUP, NOTIFICATION-GROUP, MODULE-COMPLIANCE, and AGENT-
   CAPABILITIES constructs; and via instances of objects defined in MIB
   modules, protocols, languages, specifications, interface types,
   hardware, and software.  For some of these uses other possibilities
   exist, e.g., OCTET STRING or enumerated INTEGER values.  The OBJECT
   IDENTIFIER type SHOULD be used instead of the alternatives when the
   set of identification values needs to be independently extensible
   without the need for a registry to provide centralized coordination.

   There exist a number of standard TCs that cater to some of the more
   common requirements for specialized OBJECT IDENTIFIER types.  In
   particular, SNMPv2-TC [RFC2579] contains the AutonomousType,
   VariablePointer, and RowPointer TCs.  When a standard TC provides the
   desired semantics, it SHOULD be used in an object's SYNTAX clause
   instead of OBJECT IDENTIFIER or an equivalent locally-defined TC.

   Note that OBJECT IDENTIFIER is a predefined ASN.1 type and MUST NOT
   be present in a module's IMPORTS statement.

4.6.1.6.  The BITS Construct

   The BITS construct is described in RFC 2578 Section 7.1.4.  It
   represents an enumeration of named bits.  The bit positions in a TC
   or object definition whose SYNTAX is of this type MUST start at 0 and
   SHOULD be contiguous.

   Note that the BITS construct is defined by the macros that use it and
   therefore MUST NOT be present in a module's IMPORTS statement.

4.6.1.7.  IpAddress

   The IpAddress type described in RFC 2578 Section 7.1.5 SHOULD NOT be
   used in new MIB modules.  The InetAddress/InetAddressType textual
   conventions [RFC4001] SHOULD be used instead.

4.6.1.8.  TimeTicks

   The TimeTicks type is described in RFC 2578 Section 7.1.8.  It
   represents the time in hundredths of a second between two epochs,
   reduced modulo 2^32.  It MUST NOT be subtyped, and the DESCRIPTION
   clause of any object or TC whose SYNTAX is of this type MUST identify
   the reference epochs.

   The TimeTicks type SHOULD NOT be used directly in definitions of
   objects that are snapshots of sysUpTime [RFC3418].  The TimeStamp TC
   [RFC2579] already conveys the desired semantics and SHOULD be used
   instead.

4.6.1.9.  TruthValue

   The TruthValue TC is defined in SNMPv2-TC [RFC2579].  It is an
   enumerated INTEGER type that assumes the values true(1) and false(2).

   This TC SHOULD be used in the SYNTAX clause of object definitions
   that require a Boolean type.  MIB modules SHOULD NOT use enumerated
   INTEGER types or define TCs that duplicate its semantics.

4.6.1.10.  Other Data Types

   There exist a number of standard TCs that cater to some of the more
   common requirements for specialized data types.  Some have been
   mentioned above, and Appendix B contains a partial list that includes
   those plus some others that are a bit more specialized.  An on-line
   version of that list, which is updated as new TCs are developed, can
   be found at http://www.ops.ietf.org/mib-common-tcs.html.

   Whenever a standard TC already conveys the desired semantics, it
   SHOULD be used in an object definition instead of the corresponding
   base type or a locally-defined TC.  This is especially true of the
   TCs defined in SNMPv2-TC [RFC2579] and SNMP-FRAMEWORK-MIB [RFC3411]
   because they are Internet Standards, and so modules that refer to
   them will not suffer delay in advancement on the standards track on
   account of such references.

   MIB module authors need to be aware that enumerated INTEGER or BITS
   TCs may in some cases be extended with additional enumerated values
   or additional bit positions.  When an imported TC that may be
   extended in this way is used to define an object that may be written
   or that serves as an index in a read-create table, then the set of
   values or bit positions that needs to be supported SHOULD be
   specified either in the object's DESCRIPTION clause or in an OBJECT
   clause in the MIB module's compliance statement(s).  This may be done
   by explicitly listing the required values or bit positions, or it may
   be done by stating that an implementation may support a subset of
   values or bit positions of its choosing.

4.6.2.  DESCRIPTION and REFERENCE Clauses

   It is hard to overemphasize the importance of an accurate and
   unambiguous DESCRIPTION clause for all objects and TCs.  The
   DESCRIPTION clause contains the instructions that implementors will
   use to implement an object, and if they are inadequate or ambiguous,
   then implementation quality will suffer.  Probably the single most
   important job of a MIB reviewer is to ensure that DESCRIPTION clauses
   are sufficiently clear and unambiguous to allow interoperable
   implementations to be created.

   A very common problem is to see an object definition for, say,
   'stdMIBPoofpoofCounter' with a DESCRIPTION clause that just says
   "Number of poofpoofs" with no indication what a 'poofpoof' is.  In
   such cases, it is strongly RECOMMENDED that there either be at least
   a minimal explanation or else a REFERENCE clause to point to the
   definition of a 'poofpoof'.

   For read-write objects (other than columns in read-create tables that
   have well-defined persistence properties), it is RECOMMENDED that the
   DESCRIPTION clause specify what happens to the value after an agent
   reboot.  Among the possibilities are that the value remains
   unchanged, that it reverts to a well-defined default value, or that
   the result is implementation-dependent.

4.6.3.  DISPLAY-HINT Clause

   The DISPLAY-HINT clause is used in a TC to provide a nonbinding hint
   to a management application as to how the value of an instance of an
   object defined with the syntax in the TC might be displayed.  Its
   presence is optional.

   Although management applications typically default to decimal format
   ("d") for integer TCs that are not enumerations and to a hexadecimal
   format ("1x:" or "1x " or "1x_") for octet string TCs when the
   DISPLAY-HINT clause is absent, it should be noted that SMIv2 does not
   actually specify any defaults.  MIB authors should be aware that a
   clear hint is provided to applications only when the DISPLAY-HINT
   clause is present.

4.6.4.  Conceptual Table Definitions

   RFC 2578 Sections 7.1.12 and 7.1.12.1 specify the rules for defining
   conceptual tables, and RFC 2578 Sections 7.7, 7.8, and 7.8.1 specify
   conceptual table indexing rules.  The following guidelines apply to
   such definitions:

   - For conceptual rows:

     - If the row is an extension of a row in some other table, then an
       AUGMENTS clause MUST be used if the relationship is one-to-one,
       and an INDEX clause MUST be used if the relationship is sparse.
       In the latter case, the INDEX clause SHOULD be identical to that
       of the original table.

     - If the row is an element of an expansion table -- that is, if
       multiple row instances correspond to a single row instance in
       some other table -- then an INDEX clause MUST be used, and the
       first-mentioned elements SHOULD be the indices of that other
       table, listed in the same order.

     - If objects external to the row are present in the INDEX clause,
       then the conceptual row's DESCRIPTION clause MUST specify how
       those objects are used in identifying instances of its columnar
       objects, and in particular MUST specify for which values of those
       index objects the conceptual row may exist.

     - Use of the IMPLIED keyword is NOT RECOMMENDED for any index
       object that may appear in the INDEX clause of an expansion table.
       Since this keyword may be associated only with the last object in
       an INDEX clause, it cannot be associated with the same index
       object in a primary table and an expansion table.  This will
       cause the sort order to be different in the primary table and any

       expansion tables.  As a consequence, an implementation will be
       unable to reuse indexing code from the primary table in expansion
       tables, and data structures meant to be extended might actually
       have to be replicated.  Designers who are tempted to use IMPLIED
       should consider that the resulting sort order rarely meets user
       expectations, particularly for strings that include both
       uppercase and lowercase letters, and it does not take the user
       language or locale into account.

   - If dynamic row creation and/or deletion by management applications
     is supported, then:

     - There SHOULD be one columnar object with a SYNTAX value of
       RowStatus [RFC2579] and a MAX-ACCESS value of read-create.  This
       object is called the status column for the conceptual row.  All
       other columnar objects MUST have a MAX-ACCESS value of read-
       create, read-only, accessible-for-notify, or not-accessible; a
       MAX-ACCESS value of read-write is not allowed.

     - There either MUST be one columnar object with a SYNTAX value of
       StorageType [RFC2579] and a MAX-ACCESS value of read-create, or
       else the row object (table entry) DESCRIPTION clause MUST specify
       what happens to dynamically-created rows after an agent restart.

     - If the agent itself may also create and/or delete rows, then the
       conditions under which this can occur MUST be clearly documented
       in the row object DESCRIPTION clause.

   - For conceptual rows that include a status column:

     - The DESCRIPTION clause of the status column MUST specify which
       columnar objects (if any) have to be set to valid values before
       the row can be activated.  If any objects in cascading tables
       have to be populated with related data before the row can be
       activated, then this MUST also be specified.

     - The DESCRIPTION clause of the status column MUST specify whether
       or not it is possible to modify other columns in the same
       conceptual row when the status value is active(1).  Note that in
       many cases it will be possible to modify some writable columns
       when the row is active but not others.  In such cases, the
       DESCRIPTION clause for each writable column SHOULD state whether
       or not that column can be modified when the row is active, and
       the DESCRIPTION clause for the status column SHOULD state that
       modifiability of other columns when the status value is active(1)
       is specified in the DESCRIPTION clauses for those columns (rather
       than listing the modifiable columns individually).

   - For conceptual rows that include a StorageType column:

     - The DESCRIPTION clause of the StorageType column MUST specify
       which read-write or read-create columnar objects in permanent(4)
       rows an agent must, at a minimum, allow to be writable.

   Note that RFC 2578 Section 7.8 requires that the lifetime of an
   instance of a conceptual row that AUGMENTS a base row must be the
   same as the corresponding instance of the base row.  It follows that
   there is no need for a RowStatus or StorageType column in an
   augmenting row if one is already present in the base row.

   Complete requirements for the RowStatus and StorageType TCs can be
   found in RFC 2579, in the DESCRIPTION clauses for those TCs.

4.6.5.  OID Values Assigned to Objects

   RFC 2578 Section 7.10 specifies the rules for assigning OBJECT
   IDENTIFIER (OID) values to OBJECT-TYPE definitions.  In particular:

   - A conceptual table MUST have exactly one subordinate object, which
     is a conceptual row.  The OID assigned to the conceptual row MUST
     be derived by appending a sub-identifier of "1" to the OID assigned
     to the conceptual table.

   - A conceptual row has as many subordinate objects as there are
     columns in the row; there MUST be at least one.  The OID assigned
     to each columnar object MUST be derived by appending a non-zero
     sub-identifier, unique within the row, to the OID assigned to the
     conceptual row.

   - A columnar or scalar object MUST NOT have any subordinate objects.

   - The last sub-identifier of an OID assigned to any object (be it
     table, row, column, or scalar) MUST NOT be equal to zero.  Note
     that sub-identifiers of intermediate nodes MAY be equal to zero.

   - The OID assigned to an object definition MUST NOT also be assigned
     to another definition that results in OID registration.  RFC 2578
     Section 3.6 lists the constructs that create OID registrations.

   Although it is not specifically required by the SMI, it is customary
   (and strongly RECOMMENDED) that object definitions not be registered
   beneath group definitions, compliance statements, capabilities
   statements, or notification definitions.  It is also customary (and
   strongly RECOMMENDED) that group definitions, compliance statements,

   capabilities statements, and notification definitions not be
   registered beneath object definitions.  See Appendix D for a
   RECOMMENDED OID assignment scheme.

4.6.6.  OID Length Limitations and Table Indexing

   As specified in RFC 2578 Section 3.5, all OIDs are limited to 128
   sub-identifiers.  While this is not likely to cause problems with
   administrative assignments, it does place some limitations on table
   indexing.  That is true because the length limitation also applies to
   OIDs for object instances, and these consist of the concatenation of
   the "base" OID assigned in the object definition plus the index
   components.  When a table has multiple indices of types such as OCTET
   STRING or OBJECT IDENTIFIER that resolve to multiple sub-identifiers,
   then the 128-sub-identifier limit can be quickly reached.

   Despite its inconvenience, the 128-sub-identifier limit is not
   something that can be ignored.  In addition to being imposed by the
   SMI, it is also imposed by the SNMP (see the last paragraph in
   Section 4.1 of RFC 3416 [RFC3416]).  It follows that any table with
   enough indexing components to violate this limit cannot be read or
   written using the SNMP and so is unusable.  Hence table design MUST
   take the 128-sub-identifier limit into account.  It is RECOMMENDED
   that all MIB documents make explicit any limitations on index
   component lengths that management software must observe.  This may be
   done either by including SIZE constraints on the index components or
   by specifying applicable constraints in the conceptual row
   DESCRIPTION clause or in the surrounding documentation.

4.7.  Notification Definitions

   RFC 2578 Section 8 specifies the rules for notification definitions.
   In particular:

   - Inaccessible objects MUST NOT appear in the OBJECTS clause.

   - For each object type mentioned in the OBJECTS clause, the
     DESCRIPTION clause MUST specify which object instance is to be
     present in the transmitted notification and MUST specify the
     information/meaning conveyed.

   - The OBJECT IDENTIFIER (OID) value assigned to each notification
     type MUST have a next-to-last sub-identifier of zero, so that it is
     possible to convert an SMIv2 notification definition into an SMIv1
     trap definition and back again without information loss (see
     [RFC3584] Section 2.1.2) and possible for a multilingual proxy
     chain to translate an SNMPv2 trap into an SNMPv1 trap and back
     again without information loss (see [RFC3584] Section 3).  In

     addition, the OID assigned to a notification definition MUST NOT
     also be assigned to another definition that results in OID
     registration.  RFC 2578 Section 3.6 lists the constructs that
     create OID registrations.

   Although it is not specifically required by the SMI, it is customary
   (and strongly RECOMMENDED) that notification definitions not be
   registered beneath group definitions, compliance statements,
   capabilities statements, or object definitions (this last is
   especially unwise, as it may result in an object instance and a
   notification definition sharing the same OID).  It is also customary
   (and strongly RECOMMENDED) that the OIDs assigned to notification
   types be leaf OIDs (i.e., that there be no OID registrations
   subordinate to a notification definition).  See Appendix D for a
   RECOMMENDED OID assignment scheme.

   In many cases, notifications will be triggered by external events,
   and sometimes it will be possible for those external events to occur
   at a sufficiently rapid rate that sending a notification for each
   occurrence would overwhelm the network.  In such cases, a mechanism
   MUST be provided for limiting the rate at which the notification can
   be generated.  A common technique is to require that the notification
   generator use throttling -- that is, to require that it generate no
   more than one notification for each event source in any given time
   interval of duration T.  The throttling period T MAY be configurable,
   in which case it is specified in a MIB object, or it MAY be fixed, in
   which case it is specified in the notification definition.  Examples
   of the fixed time interval technique can be found in the SNMP-
   REPEATER-MIB [RFC2108] and in the ENTITY-MIB [RFC4133].

4.8.  Compliance Statements

   RFC 2580 Sections 3, 4, and 5 specify the rules for conformance
   groups and compliance statements.  In particular:

   - Every object with a MAX-ACCESS value other than "not-accessible"
     MUST be contained in at least one object group.

   - Every notification MUST be contained in at least one notification
     group.

   - There MUST be at least one compliance statement defined for each
     "standard" MIB module.  It may reside either within that MIB module
     or within a companion MIB module.

   In writing compliance statements, there are several points that are
   easily overlooked:

   - An object group or notification group that is not mentioned either
     in the MANDATORY-GROUPS clause or in any GROUP clause of a MODULE-
     COMPLIANCE statement is unconditionally optional with respect to
     that compliance statement.  An alternate way to indicate that an
     object group or notification group is optional is to mention it in
     a GROUP clause whose DESCRIPTION clause states that the group is
     optional.  The latter method is RECOMMENDED (for optional groups
     that are relevant to the compliance statement) in order to make it
     clear that the optional status is intended rather than being the
     result of an act of omission.

   - If there are any objects with a MAX-ACCESS value of read-write or
     read-create for which there is no OBJECT clause that specifies a
     MIN-ACCESS of read-only, then implementations must support write
     access to those objects in order to be compliant with that MODULE-
     COMPLIANCE statement.  This fact sometimes catches MIB module
     authors by surprise.  When confronted with such cases, reviewers
     SHOULD verify that this is indeed what the authors intended, since
     it often is not.

   - On the other side of the coin, MIB module authors need to be aware
     that while a read-only compliance statement is sufficient to
     support interoperable monitoring applications, it is not sufficient
     to support interoperable configuration applications.  A technique
     commonly used in MIB modules that are intended to support both
     monitoring and configuration is to provide both a read-only
     compliance statement and a full compliance statement.  A good
     example is provided by the DIFFSERV-MIB [RFC3289].  Authors SHOULD
     consider using this technique when it is applicable.

   Sometimes MIB module authors will want to specify that a compliant
   implementation needs to support only a subset of the values allowed
   by an object's SYNTAX clause.  For accessible objects, this may be
   done either by specifying the required values in an object's
   DESCRIPTION clause or by providing an OBJECT clause with a refined
   SYNTAX in a compliance statement.  The latter method is RECOMMENDED
   for most cases, and is REQUIRED if there are multiple compliance
   statements with different value subsets required.  The DIFFSERV-MIB
   [RFC3289] illustrates this point.  The diffServMIBFullCompliance
   statement contains the following OBJECT clause.  (See Section 4.8.1,
   "Note Regarding These Examples and RFC 2578".)

    OBJECT       diffServDataPathStatus
    SYNTAX       RowStatus { active(1) }
    WRITE-SYNTAX RowStatus { createAndGo(4), destroy(6) }
    DESCRIPTION
       "Support for createAndWait and notInService is not required."

   whereas the diffServMIBReadOnlyCompliance statement contains this:

    OBJECT       diffServDataPathStatus
    SYNTAX       RowStatus { active(1) }
    MIN-ACCESS   read-only
    DESCRIPTION
       "Write access is not required, and active is the only status that
       needs to be supported."

   One cannot do this for inaccessible index objects because they cannot
   be present in object groups and cannot be mentioned in OBJECT
   clauses.  There are situations, however, in which one might wish to
   indicate that an implementation is required to support only a subset
   of the possible values of some index in a read-create table.  In such
   cases, the requirements MUST be specified either in the index
   object's DESCRIPTION clause (RECOMMENDED if there is only one value
   subset) or in the DESCRIPTION clause of a MODULE-COMPLIANCE statement
   (REQUIRED if the value subset is unique to the compliance statement).

   In many cases, a MIB module is always implemented in conjunction with
   one or more other MIB modules.  That fact is REQUIRED to be noted in
   the surrounding documentation (see Section 3.2 above), and it SHOULD
   also be noted in the relevant compliance statements.  In cases where
   a particular compliance statement in (say) MIB module A requires the
   complete implementation of some other MIB module B, then the
   RECOMMENDED approach is to include a statement to that effect in the
   DESCRIPTION clause of the compliance statement(s) in MIB module A.
   It is also possible, however, that MIB module A might have
   requirements that are different from those that are expressed by any
   compliance statement of module B -- for example, module A might not
   require any of the unconditionally mandatory object groups from
   module B but might require mandatory implementation of an object
   group from module B that is only conditionally mandatory with respect
   to the compliance statement(s) in module B.  In such cases, the
   RECOMMENDED approach is for the compliance statement(s) in module A
   to formally specify requirements with respect to module B via
   appropriate MODULE, MANDATORY-GROUPS, GROUP, and OBJECT clauses.  An
   example is provided by the compliance statements in the DIFFSERV-MIB
   [RFC3289], which list the ifCounterDiscontinuityGroup from IF-MIB
   [RFC2863] as a mandatory group.  That group is not sufficient to
   satisfy any IF-MIB compliance statement, and it is conditionally
   mandatory in the IF-MIB's current compliance statement ifCompliance3.

4.8.1.  Note Regarding These Examples and RFC 2578

   There has been some dispute as to whether syntax refinements that
   restrict enumerations (RFC 2578 Section 9) are permitted with TCs, as
   shown in the examples above, or are allowed only with the base types

   INTEGER and BITS, as suggested by a strict reading of RFC 2578.  The
   rough consensus of the editors of the SMIv2 documents and the current
   pool of MIB reviewers is that they should be allowed with TCs.  MIB
   module authors should be aware that some MIB compilers follow the
   strict reading of RFC 2578 and require that the TC be replaced by its
   base type (INTEGER or BITS) when enumerations are refined.  That
   usage is legal, and it can be found in some older MIB modules such as
   the IF-MIB [RFC2863].

4.9.  Revisions to MIB Modules

   RFC 2578 Section 10 specifies general rules that apply any time a MIB
   module is revised.  Specifically:

   - The MODULE-IDENTITY invocation MUST be updated to include
     information about the revision.  In particular, the LAST-UPDATED
     clause value MUST be set to the revision time, a REVISION clause
     with the same UTC time and an associated DESCRIPTION clause
     describing the changes MUST be added, and any obsolete information
     in the existing DESCRIPTION, ORGANIZATION, and CONTACT-INFO clauses
     MUST be replaced with up-to-date information.  See Section 4.5
     above for additional requirements that apply to MIB modules that
     are under IETF change control.

   - On the other hand, the module name MUST NOT be changed (except to
     correct typographical errors), existing definitions (even obsolete
     ones) MUST NOT be removed from the MIB module, and descriptors and
     OBJECT IDENTIFIER values associated with existing definitions MUST
     NOT be changed or re-assigned.

   It is important to note that the purpose in forbidding certain kinds
   of changes is to ensure that a revised MIB module is compatible with
   fielded implementations based on previous versions of the module.
   There are two distinct aspects of this backward-compatibility
   requirement.  One is "over the wire" compatibility of agent and
   manager implementations that are based on different revisions of the
   MIB module.  The other is "compilation" compatibility with MIB
   modules that import definitions from the revised MIB module.  The
   rules forbidding changing or re-assigning OBJECT IDENTIFIER values
   are necessary to ensure "over the wire" compatibility; the rules
   against changing module names or descriptors or removing obsolete
   definitions are necessary to ensure compilation compatibility.

   RFC 2578 Section 10.2 specifies rules that apply to revisions of
   object definitions.  The following guidelines correct some errors in
   these rules and provide some clarifications:

   - Bullet (1) allows the labels of named numbers and named bits in
     SYNTAX clauses of type enumerated INTEGER or BITS to be changed.
     This can break compilation compatibility, since those labels may be
     used by DEFVAL clauses in modules that import the definitions of
     the affected objects.  Therefore, labels of named numbers and named
     bits MUST NOT be changed when revising IETF MIB modules (except to
     correct typographical errors), and they SHOULD NOT be changed when
     revising enterprise MIB modules.

   - Although not specifically permitted in bullets (1) through (8), it
     is generally considered acceptable to add range constraints to the
     SYNTAX clause of an integer-valued object, provided that the
     constraints simply make explicit some value restrictions that were
     implicit in the definition of the object.  The most common example
     is an auxiliary object with a SYNTAX of INTEGER or Integer32 with
     no range constraint.  Since an auxiliary object is not permitted to
     assume negative values, adding the range constraint (0..2147483647)
     cannot possibly result in any "over the wire" change, nor will it
     cause any compilation compatibility problems with a correctly
     written MIB module.  Such a change SHOULD be treated by a reviewer
     as an editorial change, not as a semantic change.  Similarly,
     removal of a range or size constraint from an object definition
     when that range or size constraint is enforced by the underlying
     data type SHOULD be treated by a reviewer as an editorial change.

   RFC 2578 Section 10.3 specifies rules that apply to revisions of
   notification definitions.  No clarifications or corrections are
   required.

   RFC 2579 Section 5 specifies rules that apply to revisions of textual
   convention definitions.  The following guideline corrects an error in
   these rules:

   - Bullet (1) allows the labels of named numbers and named bits in
     SYNTAX clauses of type enumerated INTEGER or BITS to be changed.
     This can break compilation compatibility, since those labels may be
     used by DEFVAL clauses in modules that import the definitions of
     the affected TCs.  Therefore, labels of named numbers and named
     bits MUST NOT be changed when revising IETF MIB modules (except to
     correct typographical errors), and they SHOULD NOT be changed when
     revising enterprise MIB modules.

   RFC 2580 Section 7.1 specifies rules that apply to revisions of
   conformance groups.  Two point are worth reiterating:

   - Objects and notifications MUST NOT be added to or removed from an
     existing object group or notification group.  Doing so could cause
     a compilation failure or (worse) a silent change in the meaning of
     a compliance statement or capabilities statement that refers to
     that group.

   - The status of a conformance group is independent of the status of
     its members.  Thus, a current group MAY refer to deprecated objects
     or notifications.  This may be desirable in certain cases, e.g., a
     set of widely-deployed objects or notifications may be deprecated
     when they are replaced by a more up-to-date set of definitions, but
     the conformance groups that contain them may remain current in
     order to encourage continued implementation of the deprecated
     objects and notifications.

   RFC 2580 Section 7.2 specifies rules that apply to revisions of
   compliance statements.  The following guidelines correct an omission
   from these rules and emphasize one important point:

   - RFC 2580 should (but does not) recommend that an OBJECT clause
     specifying support for the original set of values be added to a
     compliance statement when an enumerated INTEGER object or a BITS
     object referenced by the compliance statement has enumerations or
     named bits added, assuming that no such clause is already present
     and that the effective MIN-ACCESS value is read-write or read-
     create.  This is necessary in order to avoid a silent change to the
     meaning of the compliance statement.  MIB module authors and
     reviewers SHOULD watch for this to ensure that such OBJECT clauses
     are added when needed.  Note that this may not always be possible
     to do, since affected compliance statements may reside in modules
     other than the one that contains the revised definition(s).

   - The status of a compliance statement is independent of the status
     of its members.  Thus, a current compliance statement MAY refer to
     deprecated object groups or notification groups.  This may be
     desirable in certain cases, e.g., a set of widely-deployed object
     or notification groups may be deprecated when they are replaced by
     a more up-to-date set of definitions, but compliance statements
     that refer to them may remain current in order to encourage
     continued implementation of the deprecated groups.

   RFC 2580 Section 7.3 specifies rules that apply to revisions of
   capabilities statements.  The following guideline corrects an
   omission from these rules:

   - RFC 2580 should (but does not) recommend that VARIATION clauses
     specifying support for the original set of values be added to a
     capabilities statement when enumerated INTEGER objects or BITS

     objects referenced by the capabilities statement have enumerations
     added, assuming that no such clauses are already present.  This is
     necessary in order to avoid a silent change to the meaning of the
     capabilities statement.

   In certain exceptional situations, the cost of strictly following the
   SMIv2 rules governing MIB module revisions may exceed the benefit.
   In such cases, the rules can be waived, but when that is done both
   the change and the justification for it MUST be thoroughly
   documented.  One example is provided by Section 3.1.5 of RFC 2863,
   which documents the semantic change that was made to ifIndex in the
   transition from MIB-II [RFC1213] to the IF-MIB [RFC2863] and provides
   a detailed justification for that change.  Another example is
   provided by the REVISION clause of the SONET-MIB [RFC2558] that
   documents raising the MAX-ACCESS of several objects to read-write
   while adding MIN-ACCESS of read-only for compatibility with the
   previous version [RFC1595].

   Authors and reviewers may find it helpful to use tools that can list
   the differences between two revisions of a MIB module.  Please see
   http://www.ops.ietf.org/mib-review-tools.html for more information.