Q-BRIDGE-MIB (RFC2674 [26]) provides a way to obtain only those rows
that have changed on or after some specified period of time has
passed.
One drawback to TimeFilter index tables is that a given row can
appear at many points in time, which artificially inflates the size
of the table when performing standard getNext or getBulk data
retrieval.
3.3.6.3. Alternate Data Delivery Mechanisms
If the amount of data to transfer is larger than current SNMP design
restrictions permit, as in the case of OCTET STRINGS (64k minus
overhead of IP/UDP header plus SNMP header plus varbind list plus
varbind encoding), consider delivery of the data via an alternate
method, such as FTP and use a MIB module to control that data
delivery process. In many cases, this problem can be avoided via
effective MIB design. In other words, object types requiring this
kind of transfer size should be used judiciously, if at all.
There are many enterprise MIB modules that provide control of the
TFTP or FTP protocol. Often the SNMP part defines what to send where
and setting an object initiates the operation (for an example, refer
to the CISCO-FTP-CLIENT-MIB, discussed in [38]).
Various approaches exist for allowing a local agent process running
within the managed node to take a template for an object instance
(for example for a set of interfaces), and adapt and apply it to all
of the actual instances within the node. This is an architecture for
one form of policy-based configuration (see [36], for example). Such
an architecture, which must be designed into the agent and some
portions of the MIB module, affords the efficiency of specifying many
copies of instance data only once, along with the execution
efficiency of distributing the application of the instance data to
the agent.
Other work is currently underway to improve efficiency for bulk SNMP
transfer operations [37]. The objective of these efforts is simply
the conveyance of more information with less overhead.
3.4. More Index Design Issues
Section 3.3.5 described considerations for table row index design as
it pertains to the synchronization of changes within sizable table
rows. This section simply considers how to specify this syntactically
and how to manage indices semantically.
In many respects, the design issues associated with indices in a MIB
module are similar to those in a database. Care must be taken during
the design phase to determine how often and what kind of information
must be set or retrieved. The next few points provide some guidance.
3.4.1. Simple Integer Indexing
When indexing tables using simple Integer32 or Unsigned32, start with
one (1) and specify the maximum range of the value. Since object
identifiers are unsigned long values, a question that arises is why
not index from zero (0) instead of one(1)?
RFC2578 [2], Section 7.7, page 28 states the following: Instances
identified by use of integer-valued objects should be numbered
starting from one (i.e., not from zero). The use of zero as a value
for an integer-valued index object type should be avoided, except in
special cases. Consider the provisions afforded by the following
textual convention from the Interfaces Group MIB module [33]:
InterfaceIndexOrZero ::= TEXTUAL-CONVENTION
DISPLAY-HINT "d"
STATUS current
DESCRIPTION
"This textual convention is an extension of the
InterfaceIndex convention. The latter defines a greater
than zero value used to identify an interface or interface
sub-layer in the managed system. This extension permits the
additional value of zero. the value zero is object-specific
and must therefore be defined as part of the description of
any object which uses this syntax. Examples of the usage of
zero might include situations where interface was unknown,
or when none or all interfaces need to be referenced."
SYNTAX Integer32 (0..2147483647)
3.4.2. Indexing with Network Addresses
There are many objects that use IPv4 addresses (SYNTAX IpAddress) as
indexes. One such table is the ipAddrTable from RFC2011 [14] IP-
MIB. This limits the usefulness of the MIB module to IPv4. To avoid
such limitations, use the addressing textual conventions INET-
ADDRESS-MIB [13] (or updates to that MIB module), which provides a
generic way to represent addresses for Internet Protocols. In using
the InetAddress textual convention in this MIB, however, pay heed to
the following advisory found in its description clause:
When this textual convention is used as the syntax of an index
object, there may be issues with the limit of 128 sub-identifiers
specified in SMIv2, STD 58. In this case, the OBJECT-TYPE
declaration MUST include a 'SIZE' clause to limit the number of
potential instance sub-identifiers.
One should consider the SMI limitation on the 128 sub-identifier
specification when using certain kinds of network address index
types. The most likely practical liability encountered in practice
has been with DNS names, which can in fact be in excess of 128 bytes.
The problem can be, of course, compounded when multiple indices of
this type are specified for a table.
3.5. Conflicting Controls
MIB module designers should avoid specifying read-write objects that
overlap in function partly or completely.
Consider the following situation where two read-write objects
partially overlap when a dot1dBasePortEntry has a corresponding
ifEntry.
The BRIDGE-MIB defines the following managed object:
dot1dStpPortEnable OBJECT-TYPE
SYNTAX INTEGER {
enabled(1),
disabled(2) }
ACCESS read-write
STATUS mandatory
DESCRIPTION
"The enabled/disabled status of the port."
REFERENCE
"IEEE 802.1D-1990: Section 4.5.5.2"
::= { dot1dStpPortEntry 4 }
The IF-MIB defines a similar managed object:
ifAdminStatus OBJECT-TYPE
SYNTAX INTEGER {
up(1), -- ready to pass packets
down(2),
testing(3) -- in some test mode
}
MAX-ACCESS read-write
STATUS current
DESCRIPTION
"The desired state of the interface. The testing(3)
state indicates that no operational packets can be
passed. When a managed system initializes, all
interfaces start with ifAdminStatus in the down(2) state.
As a result of either explicit management action or per
configuration information retained by the managed system,
ifAdminStatus is then changed to either the up(1) or
testing(3) states (or remains in the down(2) state)."
::= { ifEntry 7 }
If ifAdminStatus is set to testing(3), the value to be returned for
dot1dStpPortEnable is not defined. Without clarification on how
these two objects interact, management implementations will have to
monitor both objects if bridging is detected and correlate behavior.
The dot1dStpPortEnable object type could have been written with more
information about the behavior of this object when values of
ifAdminStatus which impact it change. For example, text could be
added that described proper return values for the dot1dStpPortEnable
object instance for each of the possible values of ifAdminStatus.
In those cases where overlap between objects is unavoidable, then as
we have just described, care should be taken in the description of
each of the objects to describe their possible interactions. In the
case of an object type defined after an incumbent object type, it is
necessary to include in the DESCRIPTION of this later object type the
details of these interactions.
3.6. Textual Convention Usage
Textual conventions should be used whenever possible to create a
consistent semantic for an oft-recurring datatype.
MIB modules often define a binary state object such as enable/disable
or on/off. Current practice is to use existing Textual Conventions
and define the read-write object in terms of a TruthValue from
SNMPv2-TC [3]. For example, the Q-BRIDGE-MIB [26] defines:
dot1dTrafficClassesEnabled OBJECT-TYPE
SYNTAX TruthValue
MAX-ACCESS read-write
STATUS current
DESCRIPTION
"The value true(1) indicates that Traffic Classes are
enabled on this bridge. When false(2), the bridge
operates with a single priority level for all traffic."
DEFVAL { true }
::= { dot1dExtBase 2 }
Textual conventions that have a reasonable chance of being reused in
other MIB modules ideally should also be defined in a separate MIB
module to facilitate sharing of such object types. For example, all
ATM MIB modules draw on the ATM-TC-MIB [39] to reference and utilize
common definitions for addressing, service class values, and the
like.
To simplify management, it is recommended that existing SNMPv2-TC
based definitions be used when possible. For example, consider the
following object definition:
acmePatioLights OBJECT-TYPE
SYNTAX INTEGER {
on(1),
off(2),
}
MAX-ACCESS read-write
STATUS current
DESCRIPTION
"Current status of outdoor lighting."
::= { acmeOutDoorElectricalEntry 3 }
This could be defined as follows using existing SNMPv2-TC TruthValue.
acmePatioLightsOn OBJECT-TYPE
SYNTAX TruthValue
MAX-ACCESS read-write
STATUS current
DESCRI2096PTION
"Current status of outdoor lighting. When set to true (1),
this means that the lights are enabled and turned on.
When set to false (2), the lights are turned off."
::= { acmeOutDoorElectricalEntry 3 }
3.7. Persistent Configuration
Many network devices have two levels of persistence with regard to
configuration data. In the first case, the configuration data sent
to the device is persistent only until changed with a subsequent
configuration operation, or the system is reinitialized. The second
level is where the data is made persistent as an inherent part of the
acceptance of the configuration information. Some configuration
shares both these properties, that is, that on acceptance of new
configuration data it is saved permanently and in memory. Neither of
these necessarily means that the data is used by the operational
code. Sometimes separate objects are required to activate this new
configuration data for use by the operational code.
However, many SNMP agents presently implement simple persistence
models, which do not reflect all the relationships of the
configuration data to the actual persistence model as described
above. Some SNMP set requests against MIB objects with MAX-ACCESS
read-write are written automatically to a persistent store. In other
cases, they are not. In some of the latter cases, enterprise MIB
objects are required in order to get standard configuration stored,
thus making it difficult for a generic application to have a
consistent effect.
There are standard conventions for saving configuration data. The
first method uses the Textual Convention known as StorageType [3]
which explicitly defines a given row's persistence requirement.
Examples include the RFC3231 [25] definition for the schedTable row
object schedStorageType of syntax StorageType, as well as similar row
objects for virtually all of the tables of the SNMP View-based Access
Control Model MIB [10].
A second method for persistence simply uses the DESCRIPTION clause to
define how instance data should persist. RFC2674 [26] explicitly
defines Dot1qVlanStaticEntry data persistence as follows:
dot1qVlanStaticTable OBJECT-TYPE
SYNTAX SEQUENCE OF Dot1qVlanStaticEntry
MAX-ACCESS not-accessible
STATUS current
DESCRIPTION
"A table containing static configuration information for
each VLAN configured into the device by (local or
network) management. All entries are permanent and will
be restored after the device is reset."
::= { dot1qVlan 3 }
The current practice is a dual persistence model where one can make
changes to run-time configuration as well as to a non-volatile
configuration read at device initialization. The DISMAN-SCHEDULE-MIB
module [25] provides an example of this practice. A row entry of its
SchedTable specifies the parameters by which an agent MIB variable
instance can be set to a specific value at some point in time and
governed by other constraints and directives. One of those is:
schedStorageType OBJECT-TYPE
SYNTAX StorageType
MAX-ACCESS read-create
STATUS current
DESCRIPTION
"This object defines whether this scheduled action is kept
in volatile storage and lost upon reboot or if this row is
backed up by non-volatile or permanent storage.
Conceptual rows having the value `permanent' must allow
write access to the columnar objects schedDescr,
schedInterval, schedContextName, schedVariable, schedValue,
and schedAdminStatus. If an implementation supports the
schedCalendarGroup, write access must be also allowed to
the columnar objects schedWeekDay, schedMonth, schedDay,
schedHour, schedMinute."
DEFVAL { volatile }
::= { schedEntry 19 }
It is important, however, to reiterate that the persistence is
ultimately controlled by the capabilities and features (with respect
to the storage model of management data) of the underlying system on
which the MIB Module agent is being implemented. This falls into
very much the same kind of issue set as, for example, the situation
where the size of data storage in the system for a Counter object
type is not the same as that in the corresponding MIB Object Type.
To generalize, the final word on the "when" and "how" of storage of
persistent data is dictated by the system and the implementor of the
agent on the system.
3.8. Configuration Sets and Activation
An essential notion for configuration of network elements with SNMP
is awareness of the difference between the set of one or more
configuration objects from the activation of those configuration
changes in the actual subsystem. That is, it often only makes sense
to activate a group of objects as a single 'transaction'.
3.8.1. Operational Activation Considerations
A MIB module design must consider the implications of the preceding
in the context of changes that will occur throughout a subsystem when
changes are activated. This is particularly true for configuration
changes that are complex. This complexity can be in terms of
configuration data or the operational ramifications of the activation
of the changes in the managed subsystem. A practical technique to
accommodate this kind of activation is the partitioning of contained
configuration sets, as it pertains to their being activated as
changes. Any complex configuration should have a master on/off
switch (MIB object type) as well as strategically placed on/off
switches that partition the activation of configuration data in the
managed subsystem. These controls play a pivotal role during the
configuration process as well as during subsequent diagnostics.
Generally, a series of set operations should not cause an agent to
activate each object, causing operational instability to be
introduced with every changed object instance. To avoid this
liability, ideally a series of Set PDUs can install the configuration
and a final set series of PDUs can activate the changes.
During diagnostic situations, certain on/off switches can be set to
localize the perceived error instead of having to remove the
configuration.
An example of such an object from the OSPF Version 2 MIB [29] is the
global ospfAdminStat:
ospfAdminStat OBJECT-TYPE
SYNTAX Status
MAX-ACCESS read-write
STATUS current
DESCRIPTION
"The administrative status of OSPF in the
router. The value 'enabled' denotes that the
OSPF Process is active on at least one interface;
'disabled' disables it on all interfaces."
::= { ospfGeneralGroup 2 }
Elsewhere in the OSPF MIB, the semantics of setting ospfAdminStat to
enabled(2) are clearly spelled out.
The Scheduling MIB [25] exposes such an object on each entry in the
scheduled actions table, along with the corresponding stats object
type (with read-only ACCESS) on the scheduled actions row instance.
This reflects a recurring basic design pattern which brings about
semantic clarity in the object type usage. A table can expose one
columnar object type which is strictly for administrative control.
When read, an instance of this object type will reflect its last set
or defaulted value. A companion operational columnar object type,
with MAX-ACCESS of read-only, provides the current state of
activation or deactivation resulting from the last set of the
administrative columnar instance. It is fully expected that these
administrative and operational columnar instances may reflect
different values over some period of time of activation latency,
which is why they are separate. Further sections display some of the
problems which can result from attempting to combine the operational
and administrative row columns into a single object type.
Note that all of this is independent of the RowStatus columnar
object, and the notion of 'activation' as it pertains to RowStatus.
A defined RowStatus object type should be strictly concerned with the
management of the table row itself (with 'activation' indicating "the
conceptual row is available for use by the managed device" [3], and
not to be confused with any operational activation semantics).
In the following example, schedAdminStatus controls activation of the
scheduled action, and schedOperStatus reports on its operational
status:
schedAdminStatus OBJECT-TYPE
SYNTAX INTEGER {
enabled(1),
disabled(2)
}
MAX-ACCESS read-create
STATUS current
DESCRIPTION
"The desired state of the schedule."
DEFVAL { disabled }
::= { schedEntry 14 }
schedOperStatus OBJECT-TYPE
SYNTAX INTEGER {
enabled(1),
disabled(2),
finished(3)
}
MAX-ACCESS read-only
STATUS current
DESCRIPTION
"The current operational state of this schedule. The state
enabled(1) indicates this entry is active and that the
scheduler will invoke actions at appropriate times. The
disabled(2) state indicates that this entry is currently
inactive and ignored by the scheduler. The finished(3)
state indicates that the schedule has ended. Schedules
in the finished(3) state are ignored by the scheduler.
A one-shot schedule enters the finished(3) state when it
deactivates itself."
::= { schedEntry 15 }
3.8.2. RowStatus and Deactivation
RowStatus objects should not be used to control
activation/deactivation of a configuration. While RowStatus looks
ideally suited for such a purpose since a management application can
set a row to active(1), then set it to notInService(2) to disable it
then make it active(1) again, there is no guarantee that the agent
won't discard the row while it is in the notInService(2) state. RFC
2579 [3], page 15 states:
The agent must detect conceptual rows that have been in either
state for an abnormally long period of time and remove them. It
is the responsibility of the DESCRIPTION clause of the status
column to indicate what an abnormally long period of time would
be.
The DISMAN-SCHEDULE-MIB's managed object schedAdminStatus
demonstrates how to separate row control from row activation.
Setting the schedAdminStatus to disabled(2) does not cause the row to
be aged out/removed from the table.
Finally, a reasonable agent implementation must consider how many
rows will be allowed to be created in the notReady/notInService state
such that resources are not exhausted by an errant application.
3.9. SET Operation Latency
Many standards track and enterprise MIB modules that contain read-
write objects assume that an agent can complete a set operation as
quickly as an agent can send back the status of the set operation to
the application.
Consider the subtle operational shortcomings in the following object.
It both reports the current state and allows a SET operation to
change to a possibly new state.
wheelRotationState OBJECT-TYPE
SYNTAX INTEGER { unknown(0),
idle(1),
spinClockwise(2),
spinCounterClockwise(3)
}
MAX-ACCESS read-write
STATUS current
DESCRIPTION
"The current state of a wheel."
::= { XXX 2 }
With the object defined, the following example represents one possible
transaction.
Time Command Generator --------> <--- Command Responder
----- ----------------- -----------------
|
A GetPDU(wheelRotationState.1.1)
|
| ResponsePDU(error-index 0,
| error-code 0)
|
B wheelRotationState.1.1 == spinClockwise(2)
|
C SetPDU(wheelRotationState.1.1 =
| spinCounterClockwise(3)
|
| ResponsePDU(error-index 0,
| error-code 0)
|
D wheelRotationState.1.1
== spinCounterClockwise(3)
|
E GetPDU(wheelRotationState.1.1)
|
F ResponsePDU(error-index 0,
| error-code 0)
|
V wheelRotationState.1.1 == spinClockwise(2)
....some time, perhaps seconds, later....
|
G GetPDU(wheelRotationState.1.1)
|
H ResponsePDU(error-index 0,
| error-code 0)
| wheelRotationState.1.1
V == spinCounterClockwise(3)
The response to the GET request at time E will often confuse
management applications that assume the state of the object should be
spinCounterClockwise(3). In reality, the wheel is slowing down in
order to come to the idle state then begin spinning counter
clockwise.
This possibility of confusing and paradoxical interactions of
administrative and operational state is inevitable when a single
object type is used to control and report on both types of state.
One common practice which we have already seen is to separate out the
desired (settable) state from current state. The objects
ifAdminStatus and ifOperStatus from RFC2863 [20] provide such an
example of the separation of objects into desired and current state.
3.9.1. Subsystem Latency, Persistence Latency, and Activation Latency
A second way latency can be introduced in SET operations is caused by
delay in agent implementations that must interact with loosely
coupled subsystems. The time it takes the instrumented system to
accept the new configuration information from the SNMP agent, process
it and 'install' the updated configuration in the system or otherwise
process the directives can often be longer than the SNMP response
timeout.
In these cases, it is desirable to provide a "current state" object
type which can be polled by the management application to determine
the state of control of the loosely coupled subsystem which was
affected by its configuration update.
More generally, some MIB objects may have high latencies associated
with changes to their values. This could be either a function of
saving the changed value to a persistent storage type, and/or
activating a subsystem that inherently has high latency as discussed
above. When defining such MIB objects, it might be wise to have the
agent process set operations in the managed subsystem as soon as the
Set PDU has been processed, and then update appropriate status
objects when the save-to- persistent storage and (if applicable)
activation has succeeded or is otherwise complete. Another approach
would be to cause a notification to be sent that indicates that the
operation has been completed.
When you describe an activation object, the DESCRIPTION clauses for
these objects should give a hint about the likely latency for the
completion of the operation. Keep in mind that from a management
software perspective (as presented in the example of schedAdminStatus
in Section 3.8.1), the combined latency of saving-to-persistence and
activation are not distinguishable when they are part of a single
operation.
3.10. Notifications and Error Reporting
For the purpose of this section, a 'notification' is as described in
the SMIv2, RFC2578 [2], by the NOTIFICATION-TYPE macro.
Notifications can be sent in either SNMPv2c [19] or SNMPv3 TRAP or
InformRequest PDUs. Given the sensitivity of configuration
information, it is recommended that configuration operations always
be performed using SNMPv3 due to its enhanced security capabilities.
InformRequest PDUs should be used in preference to TRAP PDUs since
the recipient of the InformRequest PDUs responds with a Response PDU.
This acknowledgment can be used to avoid unnecessary retransmission
of NOTIFICATION-TYPE information when retransmissions are in fact
required. The use of InformRequest PDUs (as opposed to TRAPs) is not
at the control of the MIB module designer or agent implementor. The
determination as to whether or not a TRAP or InformRequest PDU is
sent from an SNMPv2c or SNMPv3 agent is generally a function of the
agent's local configuration (but can be controlled with MIB objects
in SNMPv3). To the extent notification timeout and retry values are
determined by local configuration parameters, care should be taken to
avoid unnecessary retransmission of InformRequest PDUs.
Configuration change and error information conveyed in InformRequest
PDUs can be an important part of an effective SNMP-based management
system. They also have the potential to be overused. This section
offers some guidance for effective definition of NOTIFICATION-TYPE
information about configuration changes that can be carried in
InformRequest PDUs. Notifications can also play a key role for all
kinds of error reporting from hardware failures to configuration and
general policy errors. These types of notifications should be
designed as described in Section 3.11 (Application Error Reporting).
3.10.1. Identifying Source of Configuration Changes
A NOTIFICATION-TYPE designed to report configuration changes should
report the identity of the management entity initiating the
configuration change. Specifically, if the entity is known to be a
SNMP command generator, the transport address and SNMP parameters as
found in table snmpTargetParamsTable from RFC3413 SNMP-TARGET-MIB
should be reported where possible. For reporting of configuration
changes outside of the SNMP domain, the applicable change mechanism
(for example, CLI vs. HTTP-based management client access) should be
reported, along with whatever notion of "user ID" of the change
initiator is applicable and available.
3.10.2. Limiting Unnecessary Transmission of Notifications
The design of event-driven synchronization models, essential to
configuration management, can use notifications as an important
enabling technique. Proper usage of notifications allows the
manager's view of the managed element's configuration to be in close
synchronization with the actual state of the configuration of the
managed element.
When designing new NOTIFICATION-TYPEs, consider how to limit the
number of notifications PDUs that will be sent with the notification
information defined in the NOTIFICATION-TYPE in response to a
configuration change or error event.
InformRequest PDUs, when compared to TRAP PDUs, have an inherent
advantage when the concern is the reduction of unnecessary messages
from the system generating the NOTIFICATION-TYPE data, when in fact
retransmission of this data is required. That is, an InformRequest
PDU is acknowledged by the receiving entity with a Response PDU. The
receipt of this response allows the entity which generated the
InformRequest PDU to verify (and record an audit entry, where such
facilities exist on the agent system) that the message was received.
As a matter of notification protocol, this receipt guarantee is not
available when using TRAP PDUs, and if it is required, must be
accomplished by the agent using some mechanism out of band to SNMP,
and usually requiring the penalty of polling.
Regardless of the specific PDUs used to convey them, one way to limit
the unnecessary generation of notifications is to include in the
NOTIFICATION-TYPE definition situations where it need not be sent. A
good example is the frDLCIStatusChange defined in FRAME-RELAY-DTE-
MIB, RFC2115 [21].
frDLCIStatusChange NOTIFICATION-TYPE
OBJECTS { frCircuitState }
STATUS current
DESCRIPTION
"This trap indicates that the indicated Virtual Circuit
has changed state. It has either been created or
invalidated, or has toggled between the active and
inactive states. If, however, the reason for the state
change is due to the DLCMI going down, per-DLCI traps
should not be generated."
::= { frameRelayTraps 1 }
There are a number of other techniques which can be used to reduce
the unwanted generation of NOTIFICATION-TYPE information. When
defining notifications, the designer can specify a number of temporal
limitations on the generation of specific instances of a
NOTIFICATION-TYPE. For example, a definition could specify that
messages will not be sent more frequently than once every 60 seconds
while the condition which led to the generation of the notification
persists. Alternately, a NOTIFICATION-TYPE DESCRIPTION clause could
provide a fixed limit on the number of messages sent over the
duration of the condition leading to sending the notification.
If NOTIFICATION-TYPE transmission is "aggregated" in some way -
bounded either temporally or by absolute system state change as
described above - the optimal design technique is to have the data
delivered with the notification reference the actual number of
underlying managed element transitions which brought about the
notification. No matter which threshold is chosen to govern the
actual transmission of NOTIFICATION-TYPEs, the idea is to describe an
aggregated event or related set of events in as few PDUs as possible.
3.10.3. Control of Notification Subsystem
There are standards track MIB modules that define objects that either
augment or overlap control of notifications. For instance, FRAME-
RELAY-DTE-MIB RFC2115 defines frTrapMaxRate and DOCS-CABLE-DEVICE-
MIB defines a set of objects in docsDevEvent that provide for rate
limiting and filtering of notifications.
In the past, agents did not have a standard means to configure a
notification generator. With the availability of the SNMP-
NOTIFICATION-MIB module in RFC3413 [9], it is strongly recommended
that the filtering functions of this MIB module be used. This MIB
facilitates the mapping of given NOTIFICATION-TYPEs and their
intended recipients.
If the mechanisms of the SNMP-NOTIFICATION-MIB are not suitable for
this application, a explanation of why they are not suitable should
be included in the DESCRIPTION clause of any replacement control
objects.
3.11. Application Error Reporting
MIB module designers should not rely on the SNMP protocol error
reporting mechanisms alone to report application layer error state
for objects that accept SET operations.
Most MIB modules that exist today provide very little detail as to
why a configuration request has failed. Often the only information
provided is via SNMP protocol errors which generally does not provide
enough information about why an agent rejected a set request.
Typically, there is an incumbent and sizable burden on the
configuration application to determine if the configuration request
failure is the result of a resource issue, a security issue, or an
application error.
Ideally, when a "badValue" error occurs for a given set request, an
application can query the agent for more details on the error. A
badValue does not necessarily mean the command generator sent bad
data. An agent could be at fault. Additional detailed diagnostic
information may aid in diagnosing conditions in the integrated
system.
Consider the requirement of conveying error information about a MIB
expression 'object' set within the DISMAN-EXPRESSION-MIB [40] that
occurs when the expression is evaluated. Clearly, none of the
available protocol errors are relevant when reporting an error
condition that occurs when an expression is evaluated. Instead, the
DISMAN-EXPRESSION-MIB provides objects to report such errors (the
expErrorTable). Instead, the expErrorTable maintains information
about errors that occur at evaluation time:
expErrorEntry OBJECT-TYPE
SYNTAX ExpErrorEntry
MAX-ACCESS not-accessible
STATUS current
DESCRIPTION
"Information about errors in processing an expression.
Entries appear in this table only when there is a matching
expExpressionEntry and then only when there has been an
error for that expression as reflected by the error codes
defined for expErrorCode."
INDEX { expExpressionOwner, expExpressionName }
More specifically, a MIB module can provide configuration
applications with information about errors on the managed device by
creating columnar object types in log tables that contain error
information particular to errors that occur on row activation.
Notifications with detailed failure information objects can also be
used to signal configuration failures. If this approach is used, the
configuration of destinations for NOTIFICATION-TYPE data generated
from configuration failures should be considered independently of the
those for other NOTIFICATION-TYPEs which are generated for other
operational reasons. In other words, in many management
environments, the network operators interested in NOTIFICATION-TYPEs
generated from configuration failures may not completely overlap with
the community of network operators interested in NOTIFICATION-TYPEs
generated from, for example, network interface failures.
3.12. Designing MIB Modules for Multiple Managers
When designing a MIB module for configuration, there are several
pertinent considerations to provide support for multiple managers.
The first is to avoid any race conditions between two or more
authorized management applications issuing SET protocol operations
spanning over more than a single PDU.
The standard textual convention document [3] defines TestAndIncr,
often called a spinlock, which is used to avoid race conditions.
A MIB module designer may explicitly define a synchronization object
of syntax TestAndIncr or may choose to rely on snmpSetSerialNo (a
global spinlock object) as defined in SNMPv2-MIB.
snmpSetSerialNo OBJECT-TYPE
SYNTAX TestAndIncr
MAX-ACCESS read-write
STATUS current
DESCRIPTION
"An advisory lock used to allow several cooperating
command generator applications to coordinate their
use of the SNMP set operation.
This object is used for coarse-grain coordination.
To achieve fine-grain coordination, one or more similar
objects might be defined within each MIB group, as
appropriate."
::= { snmpSet 1 }
Another prominent TestAndIncr example can be found in the SNMP-
TARGET- MIB [9], snmpTargetSpinLock.
Secondly, an agent should be able to report configuration as set by
different entities as distinguishable from configuration defined
external to the SNMP domain, such as application of a default or
through an alternate management interface like a command line
interface. Section 3.10.1 describes considerations for this practice
when designing NOTIFICATION-TYPEs. The OwnerString textual
convention from RMON-MIB RFC2819 [32] has been used successfully for
this purpose. More recently, RFC3411 [1] introduced the
SnmpAdminString which has been designed as a UTF8 string. This is
more suitable for representing names in many languages.
Experience has shown that usage of OwnerString to represent row
ownership can be a useful diagnostic tool as well. Specifically, the
use of the string "monitor" to identify configuration set by an
agent/local management has been prevalent and useful in applications.
Thirdly, consider whether there is a need for multiple managers to
configure the same set of tables. If so, an "OwnerString" may be
used as the first component of a table's index to allow VACM to be
used to protect access to subsets of rows, at least at the level of
securityName or groupName provided. RFC3231 [25], Section 6
presents this technique in detail. This technique does add
complexity to the managed device and to the configuration management
application since the manager will need to be aware of these
additional columnar objects in configuration tables and act
appropriately to set them. Additionally, the agent must be
configured to provide the appropriate instance-level restrictions on
the modifiability of the instances.
3.13. Other MIB Module Design Issues
3.13.1. Octet String Aggregations
The OCTET STRING syntax can be used as an extremely flexible and
useful datatype when defining managed objects that allow SET
operation. An octet string is capable of modeling many things and is
limited in size to 65535 octets by SMIv2[2].
Since OCTET STRINGS are very flexible, the need to make them useful
to applications requires careful definition. Otherwise, applications
will at most simply be able to display and set them.
Consider the following object from RFC3418 SNMPv2-MIB [11].
sysLocation OBJECT-TYPE
SYNTAX DisplayString (SIZE (0..255))
MAX-ACCESS read-write
STATUS current
DESCRIPTION
"The physical location of this node (e.g., `telephone
closet, 3rd floor'). If the location is unknown, the value
is the zero-length string."
::= { system 6 }
Such informational object types have come to be colloquially known as
"scratch pad objects". While often useful, should an application be
required to do more with this information than be able to read and
set the value of this object, a more precise definition of the
contents of the OCTET STRING is needed, since the actual format of an
instance for such an object is unstructured. Hence, alternatively,
dividing the object type into several object type definitions can
provide the required additional structural detail.
When using OCTET STRINGS, avoid platform dependent data formats.
Also avoid using OCTET STRINGS where a more precise SMI syntax such
as SnmpAdminString or BITS would work.
There are many MIB modules that attempt to optimize the amount of
data sent/received in a SET/GET PDU by packing octet strings with
aggregate data. For example, the PortList syntax as defined in the
Q-BRIDGE-MIB (RFC2674 [26]) is defined as follows:
PortList ::= TEXTUAL-CONVENTION
STATUS current
DESCRIPTION
"Each octet within this value specifies a set of eight
ports, with the first octet specifying ports 1 through
8, the second octet specifying ports 9 through 16, etc.
Within each octet, the most significant bit represents
the lowest numbered port, and the least significant bit
represents the highest numbered port. Thus, each port
of the bridge is represented by a single bit within the
value of this object. If that bit has a value of '1'
then that port is included in the set of ports; the port
is not included if its bit has a value of '0'."
SYNTAX OCTET STRING
This compact representation saves on data transfer but has some
limitations. Such complex instance information is difficult to
reference outside of the object or use as an index to a table.
Additionally, with this approach, if a value within the aggregate
requires change, the entire aggregated object instance must be
written.
Providing an SNMP table to represent aggregate data avoids the
limitations of encoding data into OCTET STRINGs and is thus the
better general practice.
Finally, as previously mentioned in Section 3.3.6.3, one should
consider the practical ramifications of instance transfer for object
types of SYNTAX OCTET STRING where they have typical instance data
requirements close to the upper boundary of SMIv2 OCTET STRING
instance encoding. Where such object types are truly necessary at
all, SNMP/UDP may not be a very scalable means of transfer and
alternatives should be explored.
3.13.2. Supporting multiple instances of a MIB Module
When defining new MIB modules, one should consider if there could
ever be multiple instances of this MIB module in a single SNMP
entity.
MIB modules exist that assume a one to many relationship, such as
MIBs for routing protocols which can accommodate multiple "processes"
of the underlying protocol and its administrative framework.
However, the majority of MIB modules assume a one-to-one relationship
between the objects found in the MIB module and how many instances
will exist on a given SNMP agent. The OSPF-MIB, IP-MIB, BRIDGE-MIB
are all examples that are defined for a single instance of the
technology.
It is clear that single instancing of these MIB modules limits
implementations that might support multiple instances of OSPF, IP
stacks or logical bridges.
In such cases, the ENTITY-MIB [RFC2737] can provide a means for
supporting the one-to-many relationship through naming scopes using
the entLogicalTable. Keep in mind, however, that there are some
drawbacks to this approach.
1) One cannot issue a PDU request that spans naming scopes. For
example, given two instances of BRIDGE-MIB active in a single
agent, one PDU cannot contain a request for dot1dBaseNumPorts from
both the first and second instances.
2) Reliance on this technique creates a dependency on the Entity MIB
for an application to be able to access multiple instances of
information.
Alternately, completely independently of the Entity MIB, multiple MIB
module instances can be scoped by different SNMP contexts. This
does, however, require the coordination of this technique with the
administrative establishment of contexts in the configured agent
system.
3.13.3. Use of Special Optional Clauses
When defining integer-based objects for read-create, read-write and
read-only semantics, using the UNITS clause is recommended in
addition to specification in the DESCRIPTION clause of any particular
details of how UNITs are to be interpreted.
The REFERENCE clause is also recommended as a way to help an
implementer track down related information on a given object. By
adding a REFERENCE clause to the specific underlying technology
document, multiple separate implementations will be more likely to
interoperate.
4. Implementing SNMP Configuration Agents
4.1. Operational Consistency
Successful deployment of SNMP configuration systems depends on
understanding the roles of MIB module design and agent design.
Both module and agent design need to be undertaken with an
understanding of how UDP/IP-based SNMP behaves. A current practice
in MIB design is to consider the idempotency of settable objects.
Idempotency basically means being able to invoke the same set
operation repeatedly but resulting in only a single activation.
Here is an example of the idempotency in action:
Manager Agent
-------- ------
Set1 (Object A, Value B) ---> receives set OK and responds
X<-------- Response PDU(OK) is dropped by
network
Manager times out
and sends again
Set2 (Object A, Value B) ---> receives set OK (does nothing),
responds
<-------- with a Response PDU(OK)
Manager receives OK
Had object A been defined in a stateful way, the set operation might
have caused the Set2 operation to fail as a result of interaction
with Set1. If the agent implementation is not aware of such a
possible situation on the second request, the agent may behave poorly
by performing the set request again rather than doing nothing.
The example above shows that all of the software that runs on a
managed element and in managed applications should be designed in
concert when possible. Particular emphasis should be placed at the
logical boundaries of the management system components in order to
ensure correct operation.
1. The first interface is between SNMP agents in managed devices and
the management applications themselves. The MIB document is a
contract between these two entities that defines expected behavior
- it is a type of API.
2. The second interface is between the agent and the instrumented
subsystem. In some cases, the instrumented subsystem will require
modification to allow for the dynamic nature of SNMP-based
configuration, control and monitoring operations. Agent
implementors must also be sensitive to the operational code and
device in order to minimize the impact of management on the
primary subsystems.
Additionally, while the SNMP protocol-level and MIB module-level
modeling of configuration operations may be idempotent and stateless
from one set operation to another, it may not be that way in the
underlying subsystem. It is possible that an agent may need to
manage this state in these subsystem architectures explicitly when it
has placed the underlying subsystem into an "intermediate" state at a
point in processing a series of SET PDUs. Alternatively, depending
on the underlying subsystem in question, the agent may be able to
buffer all of the configuration set operations prior to activating
them in the subsystem all at once (to accommodate the nature of the
subsystem).
As an example, it would be reasonable to define a MIB module to
control Virtual Private Network (VPN) forwarding, in which a
management station could set a set of ingress/egress IP addresses for
the VPN gateway. Perhaps the MIB module presumes that the level of
transactionality is the establishment of a single row in a table
defining the address of the ingress/egress gateway, along with some
prefix information to assist in routing at the VPN layer to that
gateway. However, it would be conceivable that in an underlying
Layer 2 VPN subsystem instrumentation, the requirement is that all
existing gateways for a VPN be deleted before a new one can be
defined--that, in other words, in order to add a new gateway, g(n),
to a VPN, gateways g(1)..g(n-1) need to be removed, and then all n
gateways reestablished with the VPN forwarding service. In this
case, one could imagine an agent which has some sort of timer to
establish a bounded window for receipt of SETs for new VPN gateways,
and to activate them in this removal-then-reestablishment of existing
and new gateways at the end of this window.
4.2. Handling Multiple Managers
Devices are often modified by multiple management entities and with
different management techniques. It is sometimes the case that an
element is managed by different organizations such as when a device
sits between administrative domains.
There are a variety of approaches that management software can use to
ensure synchronization of information between the manager(s) and the
managed elements.
An agent should report configuration changes performed by different
entities. It should also distinguish configuration defined locally
such as a default or locally specified configuration made through an
alternate management interface such as a command line interface.
When a change has been made to the system via SNMP, CLI, or other
method, a managed element should send an notification to the
manager(s) configured as recipients of these applicable
notifications. These management applications should update their
local configuration repositories and then take whatever additional
action is appropriate. This approach can also be an early warning of