Francois Le Faucheur, Editor
Cisco Systems, Inc.
Waisum Lai, Co-editor
AT&T
IETF Internet Draft
Expires: March, 2003
Document: draft-ietf-tewg-diff-te-reqts-06.txt September 2002
Requirements for support of
Diff-Serv-aware MPLS Traffic Engineering
Status of this Memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026. Internet-Drafts are
Working documents of the Internet Engineering Task Force (IETF), its
areas, and its working groups. Note that other groups may also
distribute working documents as Internet-Drafts.
Internet-Drafts are draft documents valid for a maximum of six
months and may be updated, replaced, or obsoleted by other documents
at any time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt.
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html.
Abstract
This document presents the Service Provider requirements for support
of Diff-Serv aware MPLS Traffic Engineering (DS-TE).
Its objective is to provide guidance for the definition, selection
and specification of a technical solution addressing these
requirements. Specification for this solution itself is outside the
scope of this document.
A problem statement is first provided. Then, the document describes
example applications scenarios identified by Service Providers where
existing MPLS Traffic Engineering mechanisms fall short and Diff-
Serv-aware Traffic Engineering is required. The detailed
requirements that need to be addressed by the technical solution are
also reviewed. Finally, the document identifies the evaluation
criteria that should be considered for selection and definition of
the technical solution.
Le Faucheur, et. al 1
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
1. Introduction
1.1. Problem Statement
Diff-Serv is becoming prominent in providing scalable network
designs supporting multiple classes of services.
In some Diff-Serv networks where optimization of transmission
resources on a network-wide basis is not sought, MPLS Traffic
Engineering (TE) mechanisms may simply not be used.
In other networks, where optimization of transmission resources is
sought, Diff-Serv mechanisms [DIFF-MPLS] need to be complemented by
existing MPLS Traffic Engineering mechanisms [TE-REQ] [ISIS-TE]
[OSPF-TE] [RSVP-TE] [CR-LDP] which operate on an aggregate basis
across all Diff-Serv classes of service. In this case, Diff-Serv and
MPLS TE both provide their respective benefits.
Where fine-grained optimization of transmission resources is sought,
it is necessary to perform traffic engineering at a per-class level
instead of an aggregate level, in order to further enhance networks
in performance and efficiency as discussed in [TEWG-FW]. By mapping
the traffic from a given Diff-Serv class of service on a separate
LSP, it allows this traffic to utilize resources available to the
given class on both shortest path and non-shortest paths and follow
paths that meet engineering constraints which are specific to the
given class. This is what we refer to as "Diff-Serv-aware Traffic
Engineering (DS-TE)".
This document focuses exclusively on the specific environments which
would benefit from DS-TE. Some examples include:
- networks where bandwidth is scarce (e.g. transcontinental
networks)
- networks with significant amounts of delay-sensitive traffic
- networks where the relative proportion of traffic across
classes of service is not uniform
This document focuses on intra-domain operation. Inter-domain
operation is not considered.
1.2. Definitions
For the convenience of the reader, relevant Diffserv ([DIFF-ARCH],
[DIFF-NEW] and [DIFF-PDB]) definitions are repeated herein.
Behavior Aggregate (BA): a collection of packets with the same
(Diff-Serv) codepoint crossing a link in a particular direction.
Le Faucheur et. al 2
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
Per-Hop-Behavior (PHB): the externally observable forwarding
behavior applied at a DS-compliant node to a Diff-Serv behavior
aggregate.
PHB Scheduling Class (PSC): A PHB group for which a common
constraint is that ordering of at least those packets belonging
to the same microflow must be preserved.
Ordered Aggregate (OA): a set of BAs that share an ordering
constraint. The set of PHBs that are applied to this set of
Behavior Aggregates constitutes a PHB scheduling class.
Traffic Aggregate (TA): a collection of packets with a codepoint
that maps to the same PHB, usually in a DS domain or some subset
of a DS domain. A traffic aggregate marked for the foo PHB is
referred to as the "foo traffic aggregate" or "foo aggregate"
interchangeably. This generalizes the concept of Behavior
Aggregate from a link to a network.
We also repeat the following definition from [TE-REQ]:
Traffic Trunk: an aggregation of traffic flows of the same class
which are placed inside a Label Switched Path.
In the context of the present document, "flows of the same class" is
to be interpreted as "flows from the same Forwarding Equivalence
Class which are to be treated equivalently from the DS-TE
perspective".
We refer to the set of TAs corresponding to the set of PHBs of a
given PSC, as a {TA}PSC. We also loosely refer to a {TA}PSC as a
Diff-Serv class of service, or class-of service.
We refer to the collection of packets which belong to a given Traffic
Aggregate and are associated with a given MPLS Forwarding Equivalence
Class (FEC) as a <FEC/TA>.
We refer to the set of <FEC/TA> whose TAs belong to a given {TA}PSC
as a <FEC/{TA}PSC>.
1.3. Mapping of traffic to LSPs
A network may have multiple Traffic Aggregates (TAs) it wishes to
service. Recalling from [DIFF-MPLS], there are several options on
how the set of <FEC/{TA}PSC> of a given FEC can be split into
Traffic Trunks for mapping onto LSPs when running MPLS Traffic
Engineering.
One option is to not split this set of <FEC/{TA}PSC> so that each
Traffic Trunk comprises traffic from all the {TA}/PSC . This option
is typically used when aggregate traffic engineering is deployed
using current MPLS TE mechanisms. In that case, all the
Le Faucheur et. al 3
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
<FEC/{TA}PSC> of a given FEC are routed collectively according to a
single shared set of constraints and will follow the same path. Note
that the LSP transporting such a Traffic Trunk is, by definition, an
E-LSP as defined in [DIFF-MPLS].
Another option is to split the different <FEC/{TA}PSC> of a given
FEC into multiple Traffic Trunks on the basis of the {TA}PSC. In
other words, traffic from a given node to another given node, is
split based on the classes of service, into multiple Traffic Trunks
which are transported over separate LSPs, which can potentially
follow a different path through the network. DS-TE precisely takes
advantage of this fact and indeed computes a separate path for each
LSP. In so doing, DS-TE can take into account the specific
requirements of the Traffic Trunk transported on each LSP (e.g.
bandwidth requirement, preemption priority). Moreover DS-TE can take
into account specific engineering constraints to be enforced for
these sets of Traffic Trunks (e.g. limit all Traffic Trunks
transporting a particular {TA}PSC to x% of link capacity). In brief,
DS-TE achieves per LSP constraint based routing with paths that
tightly match the specific objectives of the traffic forming the
corresponding Traffic Trunk.
For simplicity, and because this is the specific topic of this
document, the above paragraphs in this section only considered
splitting traffic of a given FEC into multiple Traffic Aggregates on
the basis of {TA}PSC. However, it must be noted that, in addition to
this, traffic from every {TA}PSC may also be split into multiple
Traffic Trunks for load balancing purposes.
2. Contributing Authors
This document was the collective work of several. The text and
content of this document was contributed by the editors and the co-
authors listed below. (The contact information for the editors
appears in Section 9, and is not repeated below.)
Martin Tatham Thomas Telkamp
BT Global Crossing
Adastral Park, Martlesham Heath, Oudkerkhof 51, 3512 GJ Utrecht
Ipswich IP5 3RE, UK The Netherlands
Phone: +44-1473-606349 Phone: +31 30 238 1250
Email: martin.tatham@bt.com Email: telkamp@gblx.net
David Cooper Jim Boyle
Global Crossing Protocol Driven Networks, Inc.
960 Hamlin Court 1381 Kildaire Farm Road #288
Sunnyvale, CA 94089, USA Cary, NC 27511, USA
Phone: (916) 415-0437 Phone: (919) 852-5160
Email: dcooper@gblx.net Email: jboyle@pdnets.com
Le Faucheur et. al 4
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
Luyuan Fang Gerald R. Ash
AT&T Labs AT&T Labs
200 Laurel Avenue 200 Laurel Avenue
Middletown, New Jersey 07748, USA Middletown, New Jersey 07748,USA
Phone: (732) 420-1921 Phone: (732) 420-4578
Email: luyuanfang@att.com Email: gash@att.com
Pete Hicks Angela Chiu
CoreExpress, Inc Celion Networks
12655 Olive Blvd, Suite 500 1 Sheila Drive, Suite 2
St. Louis, MO 63141, USA Tinton Falls, NJ 07724, USA
Phone: (314) 317-7504 Phone: (732) 747-9987
Email: pete.hicks@coreexpress.net Email: angela.chiu@celion.com
William Townsend Thomas D. Nadeau
Tenor Networks Cisco Systems, Inc.
100 Nagog Park 250 Apollo Drive
Acton, MA 01720, USA Chelmsford, MA 01824, USA
Phone: +1 978-264-4900 Phone: (978) 244-3051
Email:btownsend@tenornetworks.com Email: tnadeau@cisco.com
Darek Skalecki
Nortel Networks
3500 Carling Ave,
Nepean K2H 8E9,
Phone: (613) 765-2252
Email: dareks@nortelnetworks.com
3. Application Scenarios
3.1. Scenario 1: Limiting Proportion of Classes on a Link
An IP/MPLS network may need to carry a significant amount of VoIP
traffic compared to its link capacity. For example, 10,000
uncompressed calls at 20ms packetization result in about 1Gbps of IP
traffic, which is significant on an OC-48c based network. In case of
topology changes such as link/node failure, VoIP traffic levels can
even approach the full bandwidth on certain links.
For delay/jitter reasons it is undesirable to carry more than a
certain percentage of VoIP traffic on any link. The rest of the
available link bandwidth can be used to route other classes
corresponding to delay/jitter insensitive traffic (e.g. Best Effort
Internet traffic). The exact determination of this "certain"
percentage is outside the scope of this requirements document.
During normal operations, the VoIP traffic should be able to preempt
other classes of traffic (if these other classes are designated as
preemptable and they have lower preemption priority),
Le Faucheur et. al 5
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
so that it will be able to use the shortest available path, only
constrained by the maximum defined link utilization ratio/percentage
of the VoIP class.
Existing TE mechanisms only allow constraint based routing of
traffic based on a single bandwidth constraint common to all
classes, which does not satisfy the needs described here. This leads
to the requirement for DS-TE to be able to enforce a different
bandwidth constraint for different classes of traffic. In the above
example, the bandwidth constraint to be enforced for VoIP traffic
may be the "certain" percentage of each link capacity, while the
bandwidth constraint to be enforced for the rest of the classes
might have their own constraints or have access to the rest of the
link capacity.
3.2. Scenario 2: Maintain relative proportion of traffic classes
Suppose an IP/MPLS network supports 3 classes of traffic. The
network administrator wants to perform Traffic Engineering to
distribute the traffic load. Assume also that proportion across
traffic classes varies significantly depending on the
source/destination POPs.
With existing TE mechanisms, the proportion of traffic from each
class on a given link will vary depending on multiple factors
including:
- in which order the different TE-LSPs are routed
- the preemption priority associated with the different TE-LSPs
- link/node failure situations
This may make it difficult or impossible for the network
administrator to configure the Diff-Serv PHBs (e.g. queue bandwidth)
to ensure that each traffic class gets the appropriate treatment.
This leads again to the requirement for DS-TE to be able to enforce
a different bandwidth constraint for different classes of traffic.
This could be used to ensure that, regardless of the order in which
tunnels are routed, regardless of their preemption priority and
regardless of the failure situation, the amount of traffic of each
class routed over a link matches the Diff-Serv scheduler
configuration on that link for the corresponding class (e.g. queue
bandwidth).
As an illustration of how DS-TE would address this scenario, the
network administrator may configure the service rate of Diff-Serv
queues to (45%,35%,20%) for classes (1,2,3) respectively. The
administrator would then split the traffic into separate Traffic
Trunks for each class and associate a bandwidth to each LSP
transporting those Traffic Trunks. The network administrator may
also want to configure preemption priorities of each LSP in order to
give highest restoration priority to the highest priority class and
medium priority to the medium class. Then DS-TE could ensure that
after a failure, class 1 traffic would be rerouted with first access
Le Faucheur et. al 6
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
at link capacity but without exceeding its service rate of 45% of
the link bandwidth. Class 2 traffic would be rerouted with second
access at the link capacity but without exceeding its allotment.
Note that where class 3 is the Best-Effort service, the requirement
on DS-TE may be to ensure that the total amount of traffic routed
across all classes does not exceed the total link capacity of 100%
(as opposed to separately limiting the amount of Best Effort traffic
to 20 even if there was little class 1 and class 2 traffic).
In this scenario, DS-TE allowed for the maintenance of a more steady
distribution of classes, even during rerouting. This relied on the
required capability of DS-TE to adjust the amount of traffic of each
class routed on a link based on the configuration of the scheduler
and the amount of bandwidth available for each class.
Alternatively, some network administrators may want to solve the
problem by having the scheduler dynamically adjusted based on the
amount of bandwidth of the LSPs admitted for each class. This is an
additional requirement on DS-TE.
3.3. Scenario 3: Guaranteed Bandwidth Services
In addition to the Best effort service, an IP/MPLS network operator
may desire to offer a point-to-point "guaranteed bandwidth" service
whereby the provider pledges to provide a given level of performance
(bandwidth/delay/loss...) end-to-end through its network from an
ingress port to and egress port. The goal is to ensure all
"guaranteed" traffic within a subscribed traffic contract, will be
delivered within stated tolerances.
One approach for deploying such "guaranteed" service involves:
- dedicating a Diff-Serv PHB (or a Diff-Serv PSC as defined in
[DIFF-NEW]) to the "guaranteed" traffic
- policing guaranteed traffic on ingress against the traffic
contract and marking the "guaranteed" packets with the
corresponding DSCP/EXP value
Where very high level of performance is targeted for the
"guaranteed" service, it may be necessary to ensure that the amount
of "guaranteed" traffic remains below a given percentage of link
capacity on every link. Where the proportion of "guaranteed" traffic
is high, constraint based routing can be used to enforce such a
constraint.
However, the network operator may also want to simultaneously
perform Traffic Engineering of the rest of the traffic (i.e. non-
guaranteed traffic) which would require that constraint based
routing is also capable of enforcing a different bandwidth
constraint, which would be less stringent than the one for
guaranteed traffic.
Le Faucheur et. al 7
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
Again, this combination of requirements can not be addressed with
existing TE mechanisms. DS-TE mechanisms allowing enforcement of a
different bandwidth constraint for guaranteed traffic and for non-
guaranteed traffic are required.
4. Detailed Requirements for DS-TE
This section specifies the functionality that the above scenarios
require out of DS-TE implementations. Actual technical protocol
mechanisms and procedures to achieve such functionality are outside
the scope of this document.
4.1. DS-TE Compatibility
While DS-TE is required in a number of situations such as the ones
described above, it is important to keep in mind that using DS-TE
may impact scalability (as discussed later in this document) and
operational practices. DS-TE should only be used when existing TE
mechanisms combined with Diff-Serv cannot address the network design
requirements. Many network operators may choose to not use DS-TE, or
to only use it in a limited scope within their network.
Thus, the DS-TE solution must be developed in such a way that:
(i) it raises no interoperability issues with existing deployed
TE mechanisms.
(ii) it allows DS-TE deployment to the required level of
granularity and scope (e.g. only in a subset of the
topology, or only for the number of classes required in the
considered network)
4.2. Class-Types
The fundamental requirement for DS-TE is to be able to enforce
different bandwidth constraints for different sets of Traffic
Trunks.
[TEWG-FW] introduces the concept of Class-Types when discussing
operations of MPLS Traffic Engineering in a Diff-Serv environment.
We refine this definition into the following:
Class-Type (CT): the set of Traffic Trunks crossing a link,
that is governed by a specific set of Bandwidth
constraints. CT is used for the purposes of link bandwidth
allocation, constraint based routing and admission control.
A given Traffic Trunk belongs to the same CT on all links.
Note that different LSPs transporting Traffic Trunks from the same
CT may be using the same or different preemption priorities as
explained in more details in section 3.4 below.
Le Faucheur et. al 8
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
Mapping of {TA}PSC to Class-Types is flexible. Different {TA}PSC can
be mapped to different CTs, multiple {TA}PSC can be mapped to the
same CT and one {TA}PSC can be mapped to multiple CTs.
For illustration purposes, let's consider the case of a network
running 4 Diff-Serv classes of services respectively based on the EF
PHB [EF], the AF1x PSC [AF], the AF2x PSC and the Default (i.e.
Best-Effort) PHB [DIFF-FIELD]. The network administrator may decide
to deploy DS-TE in the following way:
o from every DS-TE Head-end to every DS-TE Tail-end, split
traffic into 4 Traffic Trunks: one for traffic of each Diff-
Serv class
o because the QoS objectives for the AF1x Traffic Trunks and
for the AF2x Traffic Trunks may be of similar nature (e.g.
both targeting low loss albeit at different levels perhaps),
the same (set of) Bandwidth Constraint(s) may be applied
collectively over the AF1x Traffic Trunks and the AF2x
Traffic Trunks. Thus, the network administrator may only
define three CTs: one for the EF Traffic Trunks, one for the
AF1x and AF2x Traffic Trunks and one for the Best Effort
Traffic Trunks.
As another example of mapping of {TA}PSC to CTs, a network operator
may split the EF traffic into two different sets of traffic trunks,
so that each set of traffic trunks is subject to different
constraints on the bandwidth it can access. In this case, two
distinct CTs are defined for EF: one for the EF traffic subject to
the first (set of) bandwidth constraint(s), the other for the EF
traffic subject to the second (set of) bandwidth constraint(s).
DS-TE must support at least 2 CTs and up to 8 CTs. Those are
referred to as CTc, 0 <= c <= MaxCT-1 = 7.
In a given network, DS-TE must not require the network administrator
to always deploy the maximum number of CTs. The network
administrator must be able to deploy only the number of CTs actually
utilized.
4.3. Bandwidth Constraints
We refer to a Bandwidth Constraint Model as the set of rules
defining:
- the maximum number of Bandwidth Constraints; and
- which CTs each Bandwidth Constraint applies to and how.
By definition of CT, each CT is assigned either a Bandwidth
Constraint, or a set of Bandwidth Constraints.
We refer to the Bandwidth Constraints as BCb, 0 <= b <= MaxBC-1
Le Faucheur et. al 9
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
Different models of Bandwidth Constraints are conceivable for
control of the CTs.
For example, a model with one separate Bandwidth Constraint per CT
could be defined. This model is defined by:
- MaxBC= MaxCT
- All LSPs supporting Traffic Trunks from CTc use no more than BCc
For illustration purposes, on a link of 100 unit of bandwidth where
three CTs are used, the network administrator might then configure
BC0=30, BC1= 50, BC2=20 such that:
- All LSPs supporting Traffic Trunks from CT0 use no more than 30
(e.g. Voice <= 30)
- All LSPs supporting Traffic Trunks from CT1 use no more than 50
(e.g. Premium Data <= 50)
- All LSPs supporting Traffic Trunks from CT2 use no more than 20
(e.g. Best Effort <= 20)
As another example, a "Russian Doll" model of Bandwidth Constraints
may be defined whereby:
- MaxBC= MaxCT
- All LSPs supporting Traffic Trunks from CTc (with b<=c<=7) use no
more than BCb
For illustration purposes, on a link of 100 units of bandwidth where
three CTs are used, the network administrator might then configure
BC0=100, BC1= 80, BC2=60 such that:
- All LSPs supporting Traffic Trunks from CT2 use no more than 60
(e.g. Voice <= 60)
- All LSPs supporting Traffic Trunks from CT1 or CT2 use no more
than 80 (e.g. Voice + Premium Data <= 80)
- All LSPs supporting Traffic Trunks from CT0 or CT1 or CT2 use no
more than 100 (e.g. Voice + Premium Data + Best Effort <= 100).
Other Bandwidth Constraints model can also be conceived. Those could
involve arbitrary relationships between BCb and CTc. Those could
also involve additional concepts such as associating minimum
reservable bandwidth to a CT.
At the time of writing this document, it is not clear whether a
single model of Bandwidth Constraints is sufficient, which one it
should be and how flexible this model really needs to be and what
are the implications on the DS-TE technical solution and its
implementations. Work is currently in progress to investigate the
performance and trade-offs of different operational aspects of
Bandwidth Constraints models. The DS-TE technical solution must
specify one default bandwidth constraint model which must be
supported by any DS-TE implementation. However, additional bandwidth
constraint models may also be specified. The purpose of such a
default model is to ensure that there is at least one common
Bandwidth Constraints model implementation across various vendors
equipment and allows for easier deployment of DS-TE. However, this
Le Faucheur et. al 10
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
does not preclude a network operator to activate different Bandwidth
Constraints models on different links in a network, if he/she wishes
to do so.
In the selection of a default model, at least the following criteria
are expected to be considered:
(1) addresses the scenarios in Section 2
(2) works well under both normal and overload conditions
(3) applies equally when preemption is either enabled or disabled
(4) minimizes signaling load processing requirements
(5) maximizes efficient use of the network
In selection criteria (2), "normal condition" means that the network
is attempting to establish a volume of DS-TE LSPs for which it is
designed; "overload condition" means that the network is attempting
to establish a volume of DS-TE LSPs beyond the one it is designed
for; "works well" means that under these conditions, the network
should be able to sustain the expected performance, e.g., under
overload it is x times worse than its normal performance [BC-MODEL].
These selection criteria will be further discussed and refined as
part of the ongoing work on BC model selection. In particular, the
applicability of criterion (5) needs to be qualified.
Regardless of the Bandwidth Constraint Model, the DS-TE solution
must allow support for up to 8 BCs.
4.4. Preemption and TE-Classes
[TEWG-FW] defines the notion of preemption and preemption priority.
DS-TE must retain full support of such preemption. However, a
network administrator preferring not to use preemption for user
traffic should be able to disable the preemption mechanisms
described below.
The preemption attributes defined in [TE-REQ] must be retained and
applicable across all Class Types. The preemption attributes of
setup priority and holding priority must retain existing semantics,
and in particular these semantics must not be affected by the
Ordered Aggregate transported by the LSP or by the LSP's Class Type.
This means that if LSP1 contends with LSP2 for resources, LSP1 may
preempt LSP2 if LSP1 has a higher set-up preemption priority (i.e.
lower numerical priority value) than LSP2's holding preemption
priority regardless of LSP1's OA/CT and LSP2's OA/CT.
We introduce the following definition:
TE-Class: A pair of:
(i) a Class-Type
(ii) a preemption priority allowed for that Class-
Type. This means that an LSP transporting a
Traffic Trunk from that Class-Type can use that
Le Faucheur et. al 11
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
preemption priority as the set-up priority, as
the holding priority or both.
Note that by definition:
- for a given Class-Type, there may be one or multiple TE-classes
using that Class-Type, each using a different preemption priority
- for a given preemption priority, there may be one or multiple TE-
Class(es) using that preemption priority, each using a different
Class-Type.
DS-TE must allow all LSPs transporting Traffic Trunks of a given
Class-Type to use the same preemption priority. In other words, DS-
TE must allow a Class-Type to be used by single TE-Class. This
effectively allows the network administrator to ensure that no
preemption happens within that Class-Type, when so desired.
As an example, the DS-TE solution must allow the network
administrator to define a Class-Type comprising a single TE-class
using preemption 0.
DS-TE must allow two LSPs transporting Traffic Trunks of the same
Class-Type to use different preemption priorities, and allow the LSP
with higher (numerically lower) set-up priority to preempt the LSP
with lower (numerically higher) holding priority when they contend
for resources. In other words, DS-TE must allow multiple TE-Classes
to be defined for a given Class-Type. This effectively allows the
network administrator to enable preemption within a Class-Type, when
so desired.
As an example, the DS-TE solution must allow the network
administrator to define a Class-Type comprising three TE-Classes;
one using preemption 0, one using preemption 1 and one using
preemption 4.
DS-TE must allow two LSPs transporting Traffic Trunks from different
Class-Types to use different preemption priorities, and allow the
LSP with higher setup priority to preempt the one with lower holding
priority when they contend for resources.
As an example, the DS-TE solution must allow the network
administrator to define two Class-Types (CT0 and CT1) each
comprising two TE-Classes where say:
-one TE-Class groups CT0 and preemption 0
-one TE-Class groups CT0 and preemption 2
-one TE-Class groups CT1 and preemption 1
-one TE-Class groups CT1 and preemption 3
The network administrator would then, in particular, be able to :
- transport a CT0 Traffic Trunk over an LSP with setup priority=0
and holding priority=0
- transport a CT0 Traffic Trunk over an LSP with setup priority=2
and holding priority=0
Le Faucheur et. al 12
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
- transport a CT1 Traffic Trunk over an LSP with setup priority=1
and holding priority=1
- transport a CT1 Traffic Trunk over an LSP with setup priority=3
and holding priority=1.
The network administrator would then, in particular, NOT be able
to :
- transport a CT0 Traffic Trunk over an LSP with setup priority=1
and holding priority=1
- transport a CT1 Traffic Trunk over an LSP with setup priority=0
and holding priority=0
DS-TE must allow two LSPs transporting Traffic Trunks from different
Class-Types to use the same preemption priority. In other words, the
DS-TE solution must allow TE-classes using different CTs to use the
same preemption priority. This effectively allows the network
administrator to ensure that no preemption happens across Class-
Types, if so desired.
As an example, the DS-TE solution must allow the network
administrator to define three Class-Types (CT0, CT1 and CT2) each
comprising one TE-Class which uses preemption 0. In that case, no
preemption will ever occur.
Since there are 8 preemption priorities and up to 8 Class-Types,
there could theoretically be up to 64 TE-Classes in a network. This
is felt to be beyond current practical requirements. The current
practical requirement is that the DS-TE solution must allow support
for up to 8 TE-classes. The DS-TE solution must allow these TE-
classes to comprise any arbitrary subset of 8 (or less) from the
(64) possible combinations of (8) Class-Types and (8) preemption
priorities.
As with existing TE, an LSP which gets preempted is torn down at
preemption time. The Head-end of the preempted LSP may then attempt
to reestablish that LSP, which involves recomputing a path by
Constraint Based Routing based on updated available bandwidth
information and then signaling for LSP establishment along the new
path. It must be noted that there may be cases where the preempted
LSP cannot be reestablished (e.g. no possible path satisfying LSP
bandwidth constraints as well as other constraints). In such cases,
the Head-end behavior is left to implementation. It may involve
periodic attempts at reestablishing the LSP, relaxing of the LSP
constraints, or other behaviors.
4.5. Mapping of Traffic to LSPs
The DS-TE solution must allow operation over E-LSPs onto which a
single <FEC/{TA}PSC> is transported.
The DS-TE solution must allow operation over L-LSPs.
Le Faucheur et. al 13
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
The DS-TE solution may allow operation over E-LSPs onto which
multiple <FEC/{TA}PSC> of a given FEC are transported, under the
condition that those multiple <FEC/{TA}PSC> can effectively be
treated by DS-TE as a single atomic traffic trunk (in particular
this means that those multiple <FEC/{TA}PSC> are routed as a whole
based on a single collective bandwidth requirement, a single
affinity attribute, a single preemption level, a single Class-Type,
...). In that case, it is also assumed that the multiple {TA}PSCs
are grouped together in a consistent manner throughout the DS-TE
domain (e.g. if <FECx/{TA}PSC1> and <FECx/{TA}PSC2> are transported
together on an E-LSP, then there will not be any L-LSP transporting
<FECy/{TA}PSC1> or <FECy/{TA}PSC2> on its own, and there will not be
any E-LSP transporting <FECz/{TA}PSC1> and/or <FECz/{TA}PSC2> with
<FECz/{TA}PSC3>).
4.6. Dynamic Adjustment of Diff-Serv PHBs
As discussed in section 2.2, DS-TE may support adjustment of Diff-
Serv PHBs parameters (e.g. queue bandwidth) based on the amount of
TE-LSPs established for each OA/Class-Type. Such dynamic adjustment
is optional. It is a local matter to the LSR and as such is outside
the scope of this specification.
Where this dynamic adjustment is supported, it must allow for
disabling via configuration (thus reverting to PHB treatment with
static scheduler configuration independent of DS-TE operations). It
may involve a number of configurable parameters which are outside
the scope of this specification. Those may include configurable
parameters controlling how scheduling resources (e.g. service rates)
need to be apportioned across multiple OAs when those belong to the
same Class-Type and are transported together on the same E-LSP.
The dynamic adjustment must take account of the performance
requirements of each class when computing required adjustments.
4.7. Overbooking
Existing TE mechanisms allow overbooking to be applied on LSPs for
Constraint Based Routing and admission control. Historically this
has been achieved in TE deployment through factoring overbooking
ratios at the time of sizing the LSP bandwidth and/or at the time of
configuring the Maximum Reservable Bandwidth on links.
DS-TE must also allow overbooking and must effectively allow
different overbooking ratios to be enforced for different CTs.
DS-TE should optionally allow the effective overbooking ratio of a
given CT to be tweaked differently in different parts of the
network.
4.8. Restoration
Le Faucheur et. al 14
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
With existing TE, restoration policies use standard priority
mechanisms such as, for example, the preemption priority to
effectively control the order/importance of LSPs for restoration
purposes.
DS-TE must ensure that similar application of the use of standard
priority mechanisms for implementation of restoration policy are not
prevented since those are expected to be required for achieving the
survivability requirements of DS-TE networks.
Further discussion of restoration requirements are presented in the
output document of the TEWG Requirements Design Team [SURVIV-REQ].
5. Solution Evaluation Criteria
A range of solutions is possible for the support of the DS-TE
requirements discussed above. For example, some solutions may
require that all current TE protocols syntax (IGP, RSVP-TE, CR-LDP)
be extended in various ways. For instance, current TE protocols
could be modified to support multiple bandwidth constraints rather
than the existing single aggregate bandwidth constraint.
Alternatively, other solutions may keep the existing TE protocols
syntax unchanged but modify their semantic to allow for the multiple
bandwidth constraints.
This section identifies the evaluation criteria that should be used
to assess potential DS-TE solutions for selection.
5.1. Satisfying detailed requirements
The solution must address all the scenarios described in section 2
and satisfy all the requirements listed in section 3.
5.2. Flexibility
- number of Class Types that can be supported, compared to
number identified in Requirements section
- number of Classes within a Class-Type
5.3. Extendibility
- how far can the solution be extended in the future if
requirements for more Class-Types are identified in the
future.
5.4. Scalability
- impact on network scalability in what is propagated,
processed, stored and computed (IGP signaling, IGP
processing, IGP database, TE-Tunnel signaling ,...).
Le Faucheur et. al 15
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
- how does scalability impact evolve with number of Class-
Types/Classes actually deployed in a network. In
particular, is it possible to keep overhead small for a
large networks which only use a small number of Class-
Types/Classes, while allowing higher number of Class-
Types/Classes in smaller networks which can bear higher
overhead)
5.5. Backward compatibility/Migration
- backward compatibility/migration with/from existing TE
mechanisms
- backward compatibility/migration when
increasing/decreasing the number of Class-Types actually
deployed in a given network.
6. Security Considerations
The solution developed to address the requirements defined in this
document must address security aspects. DS-TE is not expected to add
specific security requirements beyond those of Diff-Serv and
existing TE. Networks which employ Diff-Serv techniques might offer
some protection between classes for denial of service attacks.
Though depending on how the technology is employed, it is possible
for some (lower scheduled) traffic to be more susceptible to traffic
anomalies (which include denial of service attacks) occurring within
other (higher scheduled) classes.
7. Acknowledgemnt
We thank David Allen for his help in aligning with up-to-date
Diff-Serv terminology.
8. Normative References
[AF] Heinanen, J et al., "Assured Forwarding PHB Group", RFC2597
[DIFF-ARCH] Blake et al., "An Architecture for Differentiated
Services", RFC2475.
[DIFF-MPLS] Le Faucheur et al, "Multi-Protocol Label Switching
(MPLS) Support of Differentiated Services", RFC3270, May 2002.
[DIFF-NEW] Grossman, " New Terminology and Clarifications for
Diffserv ", RFC3260, April 2002.
[EF] Davie et al., "An Expedited Forwarding PHB (Per-Hop Behavior)",
RFC3246, March 2002.
Le Faucheur et. al 16
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
[TEWG-FW] Awduche et al, Overview and Principles of Internet Traffic
Engineering, RFC3272, May 2002.
[TE-REQ] Awduche et al, Requirements for Traffic Engineering over
MPLS, RFC2702, September 1999.
9. Informative References
[CR-LDP] Jamoussi et al., "Constraint-Based LSP Setup using LDP",
RFC3212, January 2002
[DIFF-FIELD] Nichols et al., "Definition of the Differentiated
Services Field (DS Field) in the IPv4 and IPv6 Headers", RFC2474.
[DIFF-PDB] Nichols et al., "Definition of Differentiated Services
Per Domain Behaviors and Rules for their Specification", RFC3086.
[ISIS-TE] Smit, Li, IS-IS extensions for Traffic Engineering, draft-
ietf-isis-traffic-04.txt, August 2001.
[OSPF-TE] Katz, Yeung, Traffic Engineering Extensions to OSPF,
draft-katz-yeung-ospf-traffic-06.txt, October 2001.
[RSVP-TE] Awduche et al, "RSVP-TE: Extensions to RSVP for LSP
Tunnels", RFC 3209, December 2001.
[SURVIV-REQ] W.S. Lai, D. McDysan, J. Boyle, M. Carlzon, R. Coltun,
T, Griffin, E. Kern, and T. Reddington, "Network Hierarchy and
Multilayer Survivability," work in progress, October 2001.
10. Editors' Address:
Francois Le Faucheur
Cisco Systems, Inc.
Village d'Entreprise Green Side - Batiment T3
400, Avenue de Roumanille
06410 Biot-Sophia Antipolis, France
Phone: +33 4 97 23 26 19
Email: flefauch@cisco.com
Wai Sum Lai
AT&T Labs
200 Laurel Avenue
Middletown, New Jersey 07748, USA
Phone: (732) 420-3712
Email: wlai@att.com
Le Faucheur et. al 17
Requirements for Diff-Serv-aware Traffic Engineering Sep 2002
Le Faucheur et. al 18