Request for Comments: 3317 Nortel Networks
Category: Informational R. Sahita
S. Hahn
Intel
K. McCloghrie
Cisco Systems
March 2003
Differentiated Services Quality of Service Policy Information Base
Status of this Memo
This memo provides information for the Internet community. It does
not specify an Internet standard of any kind. Distribution of this
memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2003). All Rights Reserved.
Abstract
This document describes a Policy Information Base (PIB) for a device
implementing the Differentiated Services Architecture. The
provisioning classes defined here provide policy control over
resources implementing the Differentiated Services Architecture.
These provisioning classes can be used with other none Differentiated
Services provisioning classes (defined in other PIBs) to provide for
a comprehensive policy controlled mapping of service requirement to
device resource capability and usage.
Table of Contents
Conventions used in this document...................................3
1. Glossary.........................................................3
2. Introduction.....................................................3
3. Relationship to the DiffServ Informal Management Model...........3
3.1. PIB Overview.................................................4
4. Structure of the PIB.............................................6
4.1. General Conventions..........................................6
4.2. DiffServ Data Paths..........................................7
4.2.1. Data Path PRC............................................7
4.3. Classifiers..................................................8
4.3.1. Classifier PRC...........................................9
4.3.2. Classifier Element PRC...................................9
4.4. Meters.......................................................9
4.4.1. Meter PRC...............................................10
4.4.2. Token-Bucket Parameter PRC..............................10
4.5. Actions.....................................................10
4.5.1. DSCP Mark Action PRC....................................11
4.6. Queueing Elements...........................................11
4.6.1. Algorithmic Dropper PRC.................................11
4.6.2. Random Dropper PRC......................................12
4.6.3. Queues and Schedulers...................................14
4.7. Specifying Device Capabilities..............................16
5. PIB Usage Example...............................................17
5.1. Data Path Example...........................................17
5.2. Classifier and Classifier Element Example...................18
5.3. Meter Example...............................................21
5.4. Action Example..............................................21
5.5. Dropper Examples............................................22
5.5.1. Tail Dropper Example....................................22
5.5.2. Single Queue Random Dropper Example.....................23
5.5.3. Multiple Queue Random Dropper Example...................23
5.6. Queue and Scheduler Example...............................26
6. Summary of the DiffServ PIB.....................................27
7. PIB Operational Overview........................................28
8. PIB Definition..................................................29
9. Acknowledgments.................................................90
10. Security Considerations........................................90
11. Intellectual Property Considerations...........................91
12. IANA Considerations............................................91
13. Normative References...........................................92
14. Authors' Addresses.............................................95
15. Full Copyright Statement.......................................96
Conventions used in this document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in [RFC2119].
1. Glossary
PRC Provisioning Class. A type of policy data. See [POLTERM].
PRI Provisioning Instance. An instance of a PRC. See [POLTERM].
PIB Policy Information Base. The database of policy information.
See [POLTERM].
PDP Policy Decision Point. See [RAP-FRAMEWORK].
PEP Policy Enforcement Point. See [RAP-FRAMEWORK].
PRID Provisioning Instance Identifier. Uniquely identifies an
instance of a PRC.
2. Introduction
[SPPI] describes a structure for specifying policy information that
can then be transmitted to a network device for the purpose of
configuring policy at that device. The model underlying this
structure is one of well-defined provisioning classes and instances
of these classes residing in a virtual information store called the
Policy Information Base (PIB).
This document specifies a set of provisioning classes specifically
for configuring QoS Policy for Differentiated Services [DSARCH].
One way to provision policy is by means of the COPS protocol [COPS],
with the extensions for provisioning [COPS-PR]. This protocol
supports multiple clients, each of which may provision policy for a
specific policy domain such as QoS. The PRCs defined in this
DiffServ QoS PIB are intended for use by the COPS-PR diffServ client
type. Furthermore, these PRCs are in addition to any other PIBs that
may be defined for the diffServ client type in the future, as well as
the PRCs defined in the Framework PIB [FR-PIB].
3. Relationship to the DiffServ Informal Management Model
This PIB is designed according to the Differentiated Services
Informal Management Model documented in [MODEL]. The model describes
the way that ingress and egress interfaces of a 'n'-port router are
modeled. It describes the configuration and management of a DiffServ
interface in terms of a Traffic Conditioning Block (TCB) which
contains, by definition, zero or more classifiers, meters, actions,
algorithmic droppers, queues and schedulers. These elements are
arranged according to the QoS policy being expressed, and are always
in that order. Traffic may be classified; classified traffic may be
metered; each stream of traffic identified by a combination of
classifiers and meters may have some set of actions performed on it;
it may have dropping algorithms applied and it may ultimately be
stored into a queue before being scheduled out to its next
destination, either onto a link or to another TCB. When the
treatment for a given packet must have any of those elements repeated
in a way that breaks the permitted sequence {classifier, meter,
action, algorithmic dropper, queue, scheduler}, this must be modeled
by cascading multiple TCBs.
The PIB represents this cascade by following the "Next" attributes of
the various elements. They indicate what the next step in DiffServ
processing will be, whether it be a classifier, meter, action,
algorithmic dropper, queue, scheduler or a decision to now forward a
packet.
The PIB models the individual elements that make up the TCBs. The
higher level concept of a TCB is not required in the parameterization
or in the linking together of the individual elements, hence it is
not used in the PIB itself and is only mentioned in the text for
relating the PIB with the [MODEL]. The actual distinguishing of
which TCB a specific element is a part of is not needed for the
instrumentation of a device to support the functionalities of
DiffServ, but it is useful for conceptual reasons. By not using the
TCB concept, this PIB allows any grouping of elements to construct
TCBs, using rules indicated by the [MODEL]. This will minimize
changes to this PIB if rules in [MODEL] change.
The notion of a Data Path is used in this PIB to indicate the
DiffServ processing a packet may experience. This Data Path is
distinguished based on the Role Combination, Capability Set, and the
Direction of the flow the packet is part of. A Data Path Table Entry
indicates the first of possibly multiple elements that will apply
DiffServ treatment to the packet.
3.1. PIB Overview
This PIB is structured based on the need to configure the sequential
DiffServ treatments being applied to a packet, and the
parameterization of these treatments. These two aspects of the
configuration are kept separate throughout the design of the PIB, and
are fulfilled using separate tables and data definitions.
In addition, the PIB includes tables describing the capabilities and
limitations of the device using a general extensible framework.
These tables are reported to the PDP and assist the PDP with the
configuration of functional elements that can be realized by the
device.
This capabilities and limitations exchange allows a single or
multiple devices to support many different variations of a functional
datapath element. Allowing diverse methods of providing a general
functional datapath element.
In this PIB, the ingress and egress portions of a router are
configured independently but in the same manner. The difference is
distinguished by an attribute in a table describing the start of the
data path. Each interface performs some or all of the following
high-level functions:
- Classify each packet according to some set of rules.
- Determine whether the data stream the packet is part of is within
or outside its metering parameters.
- Perform a set of resulting actions such as counting and marking of
the traffic with a Differentiated Services Code Point (DSCP) as
defined in [DSFIELD].
- Apply the appropriate drop policy, either simple or complex
algorithmic drop functionality.
- Enqueue the traffic for output in the appropriate queue, whose
scheduler may shape the traffic or simply forward it with some
minimum rate or maximum latency.
The PIB therefore contains the following elements:
Data Path Table
This describes the starting point of DiffServ data paths within a
single DiffServ device. This class describes interface role
combination and interface direction specific data paths.
Classifier Tables
A general extensible framework for specifying a group of filters.
Meter Tables
A general extensible framework and one example of a
parameterization table - TBParam table, applicable for Simple
Token Bucket Meter, Average Rate Meter, Single Rate Three Color
Meter, Two Rate Three Color Meter, and Sliding Window Three Color
Meter.
Action Tables
A general extensible framework and example of parameterization
tables for Mark action. The "multiplexer" and "null" actions
described in [MODEL] are accomplished implicitly by means of the
Prid structures of the other elements.
Algorithmic Dropper Tables
A general extensible framework for describing the dropper
functional datapath element. This includes the absolute dropper
and other queue measurement dependent algorithmic droppers.
Queue and Scheduler Tables
A general extensible framework for parameterizing queuing and
scheduler systems. Notice Shaper is considered as a type of
scheduler and is included here.
Capabilities Tables
A general extensible framework for defining the capabilities and
limitations of the elements listed above. The capability tables
allow intelligent configuration of the elements by a PDP.
4. Structure of the PIB
4.1. General Conventions
The PIB consists of PRCs that represent functional elements in the
data path (e.g., classifiers, meters, actions), and classes that
specify parameters that apply to a certain type of functional element
(e.g., a Token Bucket meter or a Mark action). Parameters are
typically specified in a separate PRC to enable the use of parameter
classes by multiple policies.
Functional element PRCs use the Prid TC (defined in [SPPI]) to
indicate indirection. A Prid is an object identifier that is used to
specify an instance of a PRC in another table. A Prid is used to
point to parameter PRC that applies to a functional element, such as
which filter should be used for a classifier element. A Prid is also
used to specify an instance of a functional element PRC that
describes what treatment should be applied next for a packet in the
data path.
Note that the use of Prids to specify parameter PRCs allows the same
functional element PRC to be extended with a number of different
types of parameter PRC's. In addition, using Prids to indicate the
next functional datapath element allows the elements to be ordered in
any way.
4.2. DiffServ Data Paths
This part of the PIB provides instrumentation for connecting the
DiffServ Functional Elements within a single DiffServ device. Please
refer to [MODEL] for discussions on the valid sequencing and grouping
of DiffServ Functional Elements. Given some basic information, e.g.,
the interface capability, role combination and direction, the first
DiffServ Functional Element is determined. Subsequent DiffServ
Functional Elements are provided by the "Next" pointer attribute of
each entry of data path tables. A description of how this "Next"
pointer is used in each table is provided in their respective
DESCRIPTION clauses.
4.2.1. Data Path PRC
The Data Path PRC provides the DiffServ treatment starting points for
all packets of this DiffServ device. Each instance of this PRC
specifies the interface capability, role combination and direction
for the packet flow. There should be at most two entries for each
instance (interface type, role combination, interface capability),
one for ingress and one for egress. Each instance provides the first
DiffServ Functional Element that each packet, at a specific interface
(identified by the roles assigned to the interface) traveling in a
specific relative direction, should experience. Notice this class is
interface specific, with the use of interface type capability set and
RoleCombination. To indicate explicitly that there are no DiffServ
treatments for a particular interface type capability set, role
combination and direction, an instance of the Data Path PRC can be
created with zeroDotZero in the dsDataPathStart attribute. This
situation can also be indicated implicitly by not supplying an
instance of a Data Path PRC for that particular interface type
capability set, role combination and direction. The
explicit/implicit selection is up to the implementation. This means
that the PEP should perform normal IP device processing when
zeroDotZero is used in the dsDataPathStart attribute, or when the
entry does not exist. Normal IP device processing will depend on the
device; for example, this can be forwarding the packet.
Based on implementation experience of network devices where data path
functional elements are implemented in separate physical processors
or application specific integrated circuits, separated by switch
fabric, it seems that more complex notions of data path are required
within the network device to correlate the different physically
separate data path functional elements. For example, ingress
processing may have determined a specific ingress flow that gets
aggregated with other ingress flows at an egress data path functional
element. Some of the information determined at the ingress data path
functional element may need to be used by the egress data path
functional element. In numerous implementations, such information
has been carried by adding it to the frame/memory block used to carry
the flow within the network device; some implementers have called
such information a "preamble" or a "frame descriptor". Different
implementations use different formats for such information.
Initially, one may think such information has implementation details
within the network device that does not need to be exposed outside of
the network device. But from Policy Control point of view, such
information will be very useful in determining network resource usage
feedback from the network device to the policy server. This is
accomplished by using the Internal Label Marker and Filter PRCs
defined in [FR-PIB].
4.3. Classifiers
The classifier and classifier element tables determine how traffic is
sorted out. They identify separable classes of traffic, by reference
to appropriate filters, which may select anything from an individual
micro-flow to aggregates identified by DSCP.
The classification is used to send these separate streams to
appropriate Meter, Action, Algorithmic Dropper, Queue and Scheduler
elements. For example, to indicate a multi-stage meter, sub-classes
of traffic may be sent to different meter stages: e.g., in an
implementation of the Assured Forwarding (AF) PHB [AF-PHB], AF11
traffic might be sent to the first meter, AF12 traffic might be sent
to the second and AF13 traffic sent to the second meter stage's out-
of-profile action.
The concept of a classifier is the same as described in [MODEL]. The
structure of the classifier and classifier element tables, is the
same as the classifier described in [MODEL]. Classifier elements
have an associated precedence order solely for the purpose of
resolving ambiguity between overlapping filters. A filter with
higher values of precedence are compared first; the order of tests
for entries of the same precedence is unimportant.
A datapath may consist of more than one classifier. There may be an
overlap of filter specification between filters of different
classifiers. The first classifier functional datapath element
encountered, as determined by the sequencing of diffserv functional
datapath elements, will be used first.
An important form of classifier is "everything else": the final stage
of the classifier i.e., the one with the lowest precedence, must be
"complete" since the result of an incomplete classifier is not
necessarily deterministic - see [MODEL] section 4.1.2.
When a classifier PRC is instantiated at the PEP, it should always
have at least one classifier element table entry, the "everything
else" classifier element, with its filter matching all IP packets.
This "everything else" classifier element should be created by the
PDP as part of the classifier setup. The PDP has full control of all
classifier PRIs instantiated at the PEP.
The definition of the actual filter to be used by the classifier is
referenced via a Prid: this enables the use of any sort of filter
table that one might wish to design, standard or proprietary. No
filters are defined in this PIB. However, standard filters for IP
packets are defined in the Framework PIB [FR-PIB].
4.3.1. Classifier PRC
Classifiers, used in various ingress and egress interfaces, are
organized by the instances of the Classifier PRC. A data path entry
points to a classifier entry. A classifier entry identifies a list
of classifier elements. A classifier element effectively includes
the filter entry, and points to a "next" classifier entry or some
other data path functional element.
4.3.2. Classifier Element PRC
Classifier elements point to the filters which identify various
classes of traffic. The separation between the "classifier element"
and the "filter" allows us to use many different kinds of filters
with the same essential semantics of "an identified set of traffic".
The traffic matching the filter corresponding to a classifier element
is given to the "next" data path functional element identified in the
classifier element.
An example of a filter that may be pointed to by a Classifier Element
PRI is the frwkIpFilter PRC, defined in [FR-PIB].
4.4. Meters
A meter, according to [MODEL] section 5, measures the rate at which
packets composing a stream of traffic pass it, compares this rate to
some set of thresholds, and produces some number (two or more) of
potential results. A given packet is said to "conform" to the meter
if, at the time the packet is being looked at, the stream appears to
be within the meter's profile. PIB syntax makes it easiest to define
this as a sequence of one or more cascaded pass/fail tests, modeled
here as if-then-else constructs. It is important to understand that
this way of modeling does not imply anything about the implementation
being "sequential": multi-rate/multi-profile meters, e.g., those
designed to support [SRTCM], [TRTCM], or [TSWTCM] can still be
modeled this way even if they, of necessity, share information
between the stages: the stages are introduced merely as a notational
convenience in order to simplify the PIB structure.
4.4.1. Meter PRC
The generic meter PRC is used as a base for all more specific forms
of meter. The definition of parameters specific to the type of meter
used is referenced via a pointer to an instance of a PRC containing
those specifics. This enables the use of any sort of specific meter
table that one might wish to design, standard or proprietary. One
specific meter table is defined in this PIB module. Other meter
tables may be defined in other PIB modules.
4.4.2. Token-Bucket Parameter PRC
This is included as an example of a common type of meter. Entries in
this class are referenced from the dsMeterSpecific attributes of
meter PRC instances. The parameters are represented by a rate
dsTBParamRate, a burst size dsTBParamBurstSize, and an interval
dsTBparamInterval. The type of meter being parameterized is
indicated by the dsTBParamType attribute. This is used to determine
how the rate, burst, and rate interval parameters are used.
Additional meter parameterization classes can be defined in other
PIBs when necessary.
4.5. Actions
Actions include "no action", "mark the traffic with a DSCP" or
"specific action". Other tasks such as "shape the traffic" or "drop
based on some algorithm" are handled in other functional datapath
elements rather than in actions. The "multiplexer", "replicator",
and "null" actions described in [MODEL] are accomplished implicitly
through various combinations of the other elements.
This PIB uses the Action PRC dsActionTable to organize one Action's
relationship with the element(s) before and after it. It allows
Actions to be cascaded to enable that multiple Actions be applied to
a single traffic stream by using each entry's dsActionNext attribute.
The dsActionNext attribute of the last action entry in the chain
points to the next element in the TCB, if any, e.g., a Queueing
element. It may also point at a next TCB.
The parameters needed for the Action element will depend on the type
of Action to be taken. Hence the PIB allows for specific Action
Tables for the different Action types. This flexibility allows
additional Actions to be specified in other PIBs and also allows for
the use of proprietary Actions without impact on those defined here.
One may consider packet dropping as an Action element. Packet
dropping is handled by the Algorithmic Dropper datapath functional
element.
4.5.1. DSCP Mark Action PRC
This Action is applied to traffic in order to mark it with a DiffServ
Codepoint (DSCP) value, specified in the dsDscpMarkActTable.
4.6. Queueing Elements
These include Algorithmic Droppers, Queues and Schedulers, which are
all inter-related in their use of queueing techniques.
4.6.1. Algorithmic Dropper PRC
Algorithmic Droppers are represented in this PIB by instances of the
Algorithmic Dropper PRC. An Algorithmic Dropper is assumed to
operate indiscriminately on all packets that are presented at its
input; all traffic separation should be done by classifiers and
meters preceding it.
Algorithmic Dropper includes many types of droppers, from the simple
always dropper to the more complex random dropper. This is indicated
by the dsAlgDropType attribute.
Algorithmic Droppers have a close relationship with queuing; each
Algorithmic Dropper Table entry contains a dsAlgDropQMeasure
attribute, indicating which queue's state affects the calculation of
the Algorithmic Dropper. Each entry also contains a dsAlgDropNext
attribute that indicates to which queue the Algorithmic Dropper sinks
its traffic.
Algorithmic Droppers may also contain a pointer to a specific detail
of the drop algorithm, dsAlgDropSpecific. This PIB defines the
detail for three drop algorithms: Tail Drop, Head Drop, and Random
Drop; other algorithms are outside the scope of this PIB module, but
the general framework is intended to allow for their inclusion via
other PIB modules.
One generally-applicable parameter of a dropper is the specification
of a queue-depth threshold at which some drop action is to start.
This is represented in this PIB, as a base attribute,
dsAlgDropQThreshold, of the Algorithmic Dropper entry. The
attribute, dsAlgDropQMeasure, specifies which queue's depth
dsAlgDropQThreshold is to be compared against.
o An Always Dropper drops every packet presented to it. This type
of dropper does not require any other parameter.
o A Tail Dropper requires the specification of a maximum queue depth
threshold: when the queue pointed at by dsAlgDropQMeasure reaches
that depth threshold, dsAlgDropQThreshold, any new traffic
arriving at the dropper is discarded. This algorithm uses only
parameters that are part of the dsAlgDropEntry.
o A Head Dropper requires the specification of a maximum queue depth
threshold: when the queue pointed at by dsAlgDropQMeasure reaches
that depth threshold, dsAlgDropQThreshold, traffic currently at
the head of the queue is discarded. This algorithm uses only
parameters that are part of the dsAlgDropEntry.
o Random Droppers are recommended as a way to control congestion, in
[QUEUEMGMT] and called for in the [AF-PHB]. Various
implementations exist, that agree on marking or dropping just
enough traffic to communicate with TCP-like protocols about
congestion avoidance, but differ markedly on their specific
parameters. This PIB attempts to offer a minimal set of controls
for any random dropper, but expects that vendors will augment the
PRC with additional controls and status in accordance with their
implementation. This algorithm requires additional parameters on
top of those in dsAlgDropEntry; these are discussed below.
A Dropper Type of other is provided for the implementation of dropper
types not defined here. When the Dropper Type is other, its full
specification will need to be provided by another PRC referenced by
dsAlgDropSpecific. A Dropper Type of Multiple Queue Random Dropper
is also provided; please reference section 5.5.3 of this document for
more details.
4.6.2. Random Dropper PRC
One example of a random dropper is a RED-like dropper. An example of
the representation chosen in this PIB for this element is shown in
Figure 1.
Random droppers often have their drop probability function described
as a plot of drop probability (P) against averaged queue length (Q).
(Qmin, Pmin) then defines the start of the characteristic plot.
Normally Pmin=0, meaning that with average queue length below Qmin,
there will be no drops. (Qmax, Pmax) defines a "knee" on the plot,
after which point the drop probability become more progressive
(greater slope). (Qclip, 1) defines the queue length at which all
packets will be dropped. Notice this is different from Tail Drop
because this uses an averaged queue length. Although it is possible
for Qclip = Qmax.
In the PIB module, dsRandomDropMinThreshBytes and
dsRandomDropMinThreshPkts represent Qmin. dsRandomDropMaxThreshBytes
and dsRandomDropMaxThreshPkts represent Qmax. dsAlgDropQThreshold
represents Qclip. dsRandomDropProbMax represents Pmax. This PIB
does not represent Pmin (assumed to be zero unless otherwise
represented).
In addition, since message memory is finite, queues generally have
some upper bound above which they are incapable of storing additional
traffic. Normally this number is equal to Qclip, specified by
dsAlgDropQThreshold.
Each random dropper specification is associated with a queue. This
allows multiple drop processes (of same or different types) to be
associated with the same queue, as different PHB implementations may
require. This also allows for sequences of multiple droppers if
necessary.
+-----------------+ +-------+
|AlgDrop | |Queue |
--->| Next ---------+-+----------------->| Next -+-->
| QMeasure -------+-+ | ... |
| QThreshold | +-------+
| Type=randomDrop | +----------------+
| Specific -------+-->|RandomDrop |
+-----------------+ | MinThreshBytes |
| MaxThreshBytes |
| ProbMax |
| Weight |
| SamplingRate |
+----------------+
Figure 1: Example Use of the RandomDropTable for Random Droppers
The calculation of a smoothed queue length may also have an important
bearing on the behavior of the dropper: parameters may include the
sampling interval or rate, and the weight of each sample. The
performance may be very sensitive to the values of these parameters
and a wide range of possible values may be required due to a wide
range of link speeds. Most algorithms include a sample weight,
represented here by dsRandomDropWeight. The availability of
dsRandomDropSamplingRate as readable is important; the information
provided by the Sampling Rate is essential to the configuration of
dsRandomDropWeight. Having the Sampling Rate be configurable is also
helpful, because as line speed increases, the ability to have queue
sampling be less frequent than packet arrival is needed. Note
however that there is ongoing research on this topic, see e.g.,
[ACTQMGMT] and [AQMROUTER].
Additional parameters may be added in an enterprise PIB module, e.g.,
by using AUGMENTS on this class, to handle aspects of random drop
algorithms that are not standardized here.
NOTE: Deterministic Droppers can be viewed as a special case of
Random Droppers with the drop probability restricted to 0 and 1.
Hence Deterministic Droppers might be described by a Random Dropper
with Pmin = 0, Pmax = 1, Qmin = Qmax = Qclip, the averaged queue
length at which dropping occurs.
4.6.3. Queues and Schedulers
The Queue PRC models simple FIFO queues, as described in [MODEL]
section 7.1.1. The Scheduler PRC allows flexibility in constructing
both simple and somewhat more complex queueing hierarchies from those
queues. Of course, since TCBs can be cascaded multiple times on an
interface, even more complex hierarchies can be constructed that way
also.
Queue PRC instances are pointed at by the "next" attributes of the
upstream elements e.g., dsMeterSucceedNext. Note that multiple
upstream elements may direct their traffic to the same Queue PRI.
For example, the Assured Forwarding PHB suggests that all traffic
marked AF11, AF12, or AF13 be placed in the same queue after
metering, without reordering. This would be represented by having
the dsMeterSucceedNext of each upstream meter point at the same Queue
PRI.
NOTE: Queue and Scheduler PRIs are for data path description; they
both use Scheduler Parameterization Table entries for diffserv
treatment parameterization.
A Queue Table entry specifies the scheduler it wants service from by
use of its Next pointer.
Each Scheduler Table entry represents the algorithm in use for
servicing the one or more queues that feed it. [MODEL] section 7.1.2
describes a scheduler with multiple inputs: this is represented in
the PIB by having the scheduling parameters be associated with each
input. In this way, sets of Queues can be grouped together as inputs
to the same Scheduler. This class serves to represent the example
scheduler described in the [MODEL]: other more complex
representations might be created outside of this PIB.
Both the Queue PRC and the Scheduler PRC use instances of the
Scheduler Parameterization PRC to specify diffserv treatment
parameterization. Scheduler Parameter PRC instances are used to
parameterize each input that feeds into a scheduler. The inputs can
be a mixture of Queue PRI's and Scheduler PRI's. Scheduler Parameter
PRI's can be used/reused by one or more Queue and/or Scheduler Table
entries.
For representing a Strict Priority scheduler, each scheduler input is
assigned a priority with respect to all the other inputs feeding the
same scheduler, with default values for the other parameters. A
higher-priority input which contains traffic that is not being
delayed for shaping will be serviced before a lower-priority input.
For Weighted Scheduling methods e.g., WFQ, WRR, the "weight" of a
given scheduler input is represented with a Minimum Service Rate
leaky-bucket profile that provides a guaranteed minimum bandwidth to
that input, if required. This is represented by a rate
dsMinRateAbsolute; the classical weight is the ratio between that
rate and the interface speed, or perhaps the ratio between that rate
and the sum of the configured rates for classes. Alternatively, the
rate may be represented by a relative value, as a fraction of the
interface's current line rate, dsMinRateRelative to assist in cases
where line rates are variable or where a higher-level policy might be
expressed in terms of fractions of network resources. The two rate
parameters are inter-related and changes in one may be reflected in
the other.
For weighted scheduling methods, one can say loosely, that WRR
focuses on meeting bandwidth sharing, without concern for relative
delay amongst the queues, where WFQ control both queue service order
and amount of traffic serviced, providing meeting bandwidth sharing
and relative delay ordering amongst the queues.
A queue or scheduled set of queues (which is an input to a scheduler)
may also be capable of acting as a non-work-conserving [MODEL]
traffic shaper: this is done by defining a Maximum Service Rate
leaky-bucket profile in order to limit the scheduler bandwidth
available to that input. This is represented by a rate
dsMaxRateAbsolute; the classical weight is the ratio between that
rate and the interface speed, or perhaps the ratio between that rate
and the sum of the configured rates for classes. Alternatively, the
rate may, be represented by a relative value, as a fraction of the
interface's current line rate, dsMaxRateRelative. There was
discussion in the working group about alternative modeling
approaches, such as defining a shaping action or a shaping element.
We did not take this approach because shaping is in fact something a
scheduler does to its inputs, (which we model as a queue with a
maximum rate or a scheduler whose output has a maximum rate) and we
felt it was simpler and more elegant to simply describe it in that
context. Additionally, multi-rate shaper [SHAPER] can be represented
by the use of multiple dsMaxRateTable entries.
Other types of priority and weighted scheduling methods can be
defined using existing parameters in dsMinRateEntry. NOTE:
dsSchedulerMethod uses AutonomousType syntax, with the different
types of scheduling methods defined as OBJECT-IDENTITY. Future
scheduling methods may be defined in other PIBs. This requires an
OBJECT-IDENTITY definition, a description of how the existing objects
are reused, if they are, and any new objects they require.
NOTE: Hierarchical schedulers can be parameterized using this PIB by
having Scheduler Table entries feeds into Scheduler Table entry.
4.7. Specifying Device Capabilities
The DiffServ PIB uses the Base PRC classes frwkPrcSupportTable and
frwkCompLimitsTable defined in [FR-PIB] to specify what PRC's are
supported by a PEP and to specify any limitations on that support.
The PIB also uses the capability PRC's frwkCapabilitySetTable and
frwkIfRoleComboTable defined in [FR-PIB] to specify the device's
capability sets, interface types, and role combinations. Each
instance of the capability PRC frwkCapabilitySetTable contains an OID
that points to an instance of a PRC that describes some capability of
that interface type. The DiffServ PIB defines several of these
capability PRCs, that assist the PDP with the configuration of
DiffServ functional elements that can be implemented by the device.
Each of these capability PRCs contains a direction attribute that
specifies the direction for which the capability applies. This
attribute is defined in a base capability PRC, which is extended by
each specific capability PRC.
Classification capabilities, which specify the information elements
the device can use to classify traffic, are reported using the
dsIfClassificationCaps PRC. Metering capabilities, which indicate
what the device can do with out-of-profile packets, are specified
using the dsIfMeteringCaps PRC. Scheduling capabilities, such as the
number of inputs supported, are reported using the dsIfSchedulingCaps
PRC. Algorithmic drop capabilities, such as the types of algorithms
supported, are reported using the dsIfAlgDropCaps PRC. Queue
capabilities, such as the maximum number of queues, are reported
using the dsIfQueueCaps PRC. Maximum Rate capabilities, such as the
maximum number of max rate Levels, are reported using the
dsIfMaxRateCaps PRC.
Two PRC's are defined to allow specification of the element linkage
capabilities of the PEP. The dsIfElmDepthCaps PRC indicates the
maximum number of functional datapath elements that can be linked
consecutively in a datapath. The dsIfElmLinkCaps PRC indicates what
functional datapath elements may follow a specific type of element in
a datapath.
The capability reporting classes in the DiffServ and Framework PIB
are meant to allow the PEP to indicate some general guidelines about
what the device can do. They are intended to be an aid to the PDP
when it constructs policy for the PEP. These classes do not
necessarily allow the PEP to indicate every possible configuration
that it can or cannot support. If a PEP receives a policy that it
cannot implement, it must notify the PDP with a failure report.
Currently [COPS-PR] error handling mechanism as specified in [COPS-
PR] sections 4.4, 4.5, and 4.6 completely handles all known error
cases of this PIB; hence no additional methods or PRCs need to be
specified here.
5. PIB Usage Example
This section provides some examples on how the different table
entries of this PIB may be used together for a DiffServ Device. The
usage of each individual attribute is defined within the PIB module
itself. For the figures, all the PIB table entry and attribute names
are assumed to have "ds" as their first common initial part of the
name, with the table entry name assumed to be their second common
initial part of the name. "0.0" is being used to mean zeroDotZero.
And for Scheduler Method "= X" means "using the OID of
diffServSchedulerX".
5.1. Data Path Example
Notice Each entry of the DataPath table is used for a specific
interface type handling a flow in a specific direction for a specific
functional role-combination. For our example, we just define one
such entry.
+---------------------+
|DataPath |
| CapSetName ="IfCap1"|
| Roles = "A+B" |
| IfDirection=Ingress | +---------+
| Start --------------+--->|Clfr |
+---------------------+ | Id=Dept |
+---------+
Figure 2: DataPath Usage Example
In Figure 2, we are using IfCap1 to indicate interface type with
capability set 1 handling ingress flow for functional roles of "A+B".
We are using classifier for departments to lead us into the
Classifier Example below.
5.2. Classifier and Classifier Element Example
We want to show how a multilevel classifier can be built using the
classifier tables provided by this PIB. Notice we didn't go into
details on the filters because they are not defined by this PIB.
Continuing in the Data Path example from the previous section, lets
say we want to perform the following classification functionality to
do flow separation based on department and application type:
if (Dept1) then take Dept1-action
{
if (Appl1) then take Dept1-Appl1-action.
if (Appl2) then take Dept1-Appl2-action.
if (Appl3) then take Dept1-Appl3-action.
}
if (Dept2) then take Dept2-action
{
if (Appl1) then take Dept2-Appl1-action.
if (Appl2) then take Dept2-Appl2-action.
if (Appl3) then take Dept2-Appl3-action.
}
if (Dept3) then take Dept3-action
{
if (Appl1) then take Dept3-Appl1-action.
if (Appl2) then take Dept3-Appl2-action.
if (Appl3) then take Dept3-Appl3-action.
}
The above classification logic is translated into the following PIB
table entries, with two levels of classifications.
First for department:
+---------+
|Clfr |
| Id=Dept |
+---------+
+-------------+ +-----------+
|ClfrElement | +-->|Clfr |
| Id=Dept1 | | | Id=D1Appl |
| ClfrId=Dept | | +-----------+
| Preced=NA | |
| Next -------+--+ +------------+
| Specific ---+----->|Filter Dept1|
+-------------+ +------------+
+-------------+ +-----------+
|ClfrElement | +-->|Clfr |
| Id=Dept2 | | | Id=D2Appl |
| ClfrId=Dept | | +-----------+
| Preced=NA | |
| Next -------+--+ +------------+
| Specific ---+----->|Filter Dept2|
+-------------+ +------------+
+-------------+ +-----------+
|ClfrElement | +-->|Clfr |
| Id=Dept3 | | | Id=D3Appl |
| ClfrId=Dept | | +-----------+
| Preced=NA | |
| Next -------+--+ +------------+
| Specific ---+----->|Filter Dept3|
+-------------+ +------------+
Second for application:
+-----------+
|Clfr |
| Id=D1Appl |
+-----------+
+---------------+ +--------------+
|ClfrElement | +----------------->|Meter |
| Id=D1Appl1 | | | Id=D1A1Rate1 |
| ClfrId=D1Appl | | | SucceedNext -+--->...
| Preced=NA | | | FailNext ----+--->...
| Next ---------+--+ +------------+ | Specific ----+--->...
| Specific -----+---->|Filter Appl1| +--------------+
+---------------+ +------------+
+---------------+ +--------------+
|ClfrElement | +----------------->|Meter |
| Id=D1Appl2 | | | Id=D1A2Rate1 |
| ClfrId=D1Appl | | | SucceedNext -+--->...
| Preced=NA | | | FailNext ----+--->...
| Next ---------+--+ +------------+ | Specific ----+--->...
| Specific -----+---->|Filter Appl2| +--------------+
+---------------+ +------------+
+---------------+ +--------------+
|ClfrElement | +----------------->|Meter |
| Id=D1Appl3 | | | Id=D1A3Rate1 |
| ClfrId=D1Appl | | | SucceedNext -+--->...
| Preced=NA | | | FailNext ----+--->...
| Next ---------+--+ +------------+ | Specific ----+--->...
| Specific -----+---->|Filter Appl3| +--------------+
+---------------+ +------------+
Figure 3: Classifier Usage Example
The application classifiers for department 2 and 3 will be very much
like the application classifier for department 1 shown above. Notice
in this example, Filters for Appl1, Appl2, and Appl3 are reusable
across the application classifiers.
This classifier and classifier element example assume the next
differentiated services functional datapath element is Meter and
leads us into the Meter Example section.
5.3. Meter Example
A single rate simple Meter may be easy to envision, hence we will do
a Two Rate Three Color [TRTCM] example, using two Meter table entries
and two TBParam table entries.
+--------------+ +---------+ +--------------+ +----------+
|Meter | +->|Action | +->| Meter | +->|Action |
| Id=D1A1Rate1 | | | Id=Green| | | Id=D1A1Rate2 | | | Id=Yellow|
| SucceedNext -+-+ +---------+ | | SucceedNext -+-+ +----------+
| FailNext ----+-----------------+ | FailNext ----+--+ +-------+
| Specific -+ | | Specific -+ | +->|Action |
+-----------+--+ +-----------+--+ | Id=Red|
| | +-------+
| +------------+ | +------------+
+->|TBParam | +->|TBParam |
| Type=TRTCM | | Type=TRTCM |
| Rate | | Rate |
| BurstSize | | BurstSize |
| Interval | | Interval |
+------------+ +------------+
Figure 4: Meter Usage Example
For [TRTCM], the first level TBParam entry is used for Committed
Information Rate and Committed Burst Size Token Bucket, and the
second level TBParam entry is used for Peak Information Rate and Peak
Burst Size Token Bucket.
The other meters needed for this example will depend on the service
class each classified flow uses. But their construction will be
similar to the example given here. The TBParam table entries can be
shared by multiple Meter table entries.
In this example the differentiated services functional datapath
element following Meter is Action, detailed in the following section.
5.4. Action Example
Typically, Mark Action will be used; we will continue using the
"Action, Id=Green" branch off the Meter example.
Recall this is the D1A1Rate1 SucceedNext branch, meaning the flow
belongs to Department 1 Application 1, within the committed rate and
burst size limits for this flow. We would like to Mark this flow
with a specific DSCP and also with a device internal label.
+-----------+ +-----------+ +--->AlgDropAF11
|Action | +----------------->|Action | |
| Next -----+--+ +------------+ | Next -----+--+ +-------------+
| Specific -+---->|DscpMarkAct | | Specific -+--->|ILabelMarker |
+-----------+ | Dscp=AF11 | +-----------+ | ILabel=D1A1 |
+------------+ +-------------+
Figure 5: Action Usage Example
This example uses the frwkILabelMarker PRC defined in [FR-PIB],
showing the device internal label being used to indicate the micro
flow that feeds into the aggregated AF flow. This device internal
label may be used for flow accounting purposes and/or other data path
treatments.
5.5. Dropper Examples
The Dropper examples below will continue from the Action example
above for AF11 flow. We will provide three different dropper setups,
from simple to complex. The examples below may include some queuing
structures; they are here only to show the relationship of the
droppers to queuing and are not complete. Queuing examples are
provided in later sections.
5.5.1. Tail Dropper Example
The Tail Dropper is one of the simplest. For this example we just
want to drop part of the flow that exceeds the queue's buffering
capacity, 2 Mbytes.
+--------------------+ +------+
|AlgDrop | +->|Q AF1 |
| Id=AF11 | | +------+
| Type=tailDrop | |
| Next --------------+-+--+
| QMeasure ----------+-+
| QThreshold=2Mbytes |
| Specific=0.0 |
+--------------------+
Figure 6: Tail Dropper Usage Example
5.5.2. Single Queue Random Dropper Example
The use of Random Dropper will introduce the usage of
dsRandomDropEntry as in the example below.
+-----------------+ +------+
|AlgDrop | +->|Q AF1 |
| Id=AF11 | | +------+
| Type=randomDrop | |
| Next -----------+-+--+
| QMeasure -------+-+
| QThreshold | +----------------+
| Specific -------+-->|RandomDrop |
+-----------------+ | MinThreshBytes |
| MinThreshPkts |
| MaxThreshBytes |
| MaxThreshPkts |
| ProbMax |
| Weight |
| SamplingRate |
+----------------+
Figure 7: Single Queue Random Dropper Usage Example
Notice for Random Dropper, dsAlgDropQThreshold contains the maximum
average queue length, Qclip, for the queue being measured as
indicated by dsAlgDropQMeasure, the rest of the Random Dropper
parameters are specified by dsRandomDropEntry as referenced by
dsAlgDropSpecific. In this example, both dsAlgDropNext and
dsAlgDropQMeasure references the same queue. This is the simple case
but dsAlgDropQMeasure may reference another queue for PEP
implementation supporting this feature.
5.5.3. Multiple Queue Random Dropper Example
When network device implementation requires measuring multiple queues
in determining the behavior of a drop algorithm, the existing PRCs
defined in this PIB will be sufficient for the simple case, as
indicated by this example.
+-------------+ +------+
|AlgDrop | +----------------+-------------------+->|Q_AF1 |
| Id=AF11 | | | | +------+
| Type=mQDrop | | | |
| Next -------+-+ +------------+ | +------------+ |
| QMeasure ---+-->|MQAlgDrop | | +->|MQAlgDrop | |