Network Working Group D. Awduche
Request for Comments: 3209 Movaz Networks, Inc.
Category: Standards Track L. BergerD. Gan
Juniper Networks, Inc.
Procket Networks, Inc.
Cosine Communications, Inc.
Cisco Systems, Inc.
December 2001 RSVP-TE: Extensions to RSVP for LSP Tunnels
Status of this Memo
This document specifies an Internet standards track protocol for the
Internet community, and requests discussion and suggestions for
improvements. Please refer to the current edition of the "Internet
Official Protocol Standards" (STD 1) for the standardization state
and status of this protocol. Distribution of this memo is unlimited.
Copyright (C) The Internet Society (2001). All Rights Reserved.
This document describes the use of RSVP (Resource Reservation
Protocol), including all the necessary extensions, to establish
label-switched paths (LSPs) in MPLS (Multi-Protocol Label Switching).
Since the flow along an LSP is completely identified by the label
applied at the ingress node of the path, these paths may be treated
as tunnels. A key application of LSP tunnels is traffic engineering
with MPLS as specified in RFC 2702.
We propose several additional objects that extend RSVP, allowing the
establishment of explicitly routed label switched paths using RSVP as
a signaling protocol. The result is the instantiation of label-
switched tunnels which can be automatically routed away from network
failures, congestion, and bottlenecks.
4.6.4 Reroute and Bandwidth Increase Procedure ............... 424.7 Session Attribute Object ............................... 434.7.1 Format without resource affinities ..................... 434.7.2 Format with resource affinities ........................ 454.7.3 Procedures applying to both C-Types .................... 464.7.4 Resource Affinity Procedures .......................... 485 Hello Extension ........................................ 495.1 Hello Message Format ................................... 505.2 HELLO Object formats ................................... 515.2.1 HELLO REQUEST object ................................... 515.2.2 HELLO ACK object ....................................... 515.3 Hello Message Usage .................................... 525.4 Multi-Link Considerations .............................. 535.5 Compatibility .......................................... 546 Security Considerations ................................ 547 IANA Considerations .................................... 547.1 Message Types .......................................... 557.2 Class Numbers and C-Types .............................. 557.3 Error Codes and Globally-Defined Error Value Sub-Codes . 577.4 Subobject Definitions .................................. 578 Intellectual Property Considerations ................... 589 Acknowledgments ........................................ 5810 References ............................................. 5811 Authors' Addresses ..................................... 6012 Full Copyright Statement ............................... 611. Introduction
Section 2.9 of the MPLS architecture  defines a label distribution
protocol as a set of procedures by which one Label Switched Router
(LSR) informs another of the meaning of labels used to forward
traffic between and through them. The MPLS architecture does not
assume a single label distribution protocol. This document is a
specification of extensions to RSVP for establishing label switched
paths (LSPs) in MPLS networks.
Several of the new features described in this document were motivated
by the requirements for traffic engineering over MPLS (see ). In
particular, the extended RSVP protocol supports the instantiation of
explicitly routed LSPs, with or without resource reservations. It
also supports smooth rerouting of LSPs, preemption, and loop
The LSPs created with RSVP can be used to carry the "Traffic Trunks"
described in . The LSP which carries a traffic trunk and a
traffic trunk are distinct though closely related concepts. For
example, two LSPs between the same source and destination could be
load shared to carry a single traffic trunk. Conversely several
traffic trunks could be carried in the same LSP if, for instance, the
LSP were capable of carrying several service classes. The
applicability of these extensions is discussed further in .
Since the traffic that flows along a label-switched path is defined
by the label applied at the ingress node of the LSP, these paths can
be treated as tunnels, tunneling below normal IP routing and
filtering mechanisms. When an LSP is used in this way we refer to it
as an LSP tunnel.
LSP tunnels allow the implementation of a variety of policies related
to network performance optimization. For example, LSP tunnels can be
automatically or manually routed away from network failures,
congestion, and bottlenecks. Furthermore, multiple parallel LSP
tunnels can be established between two nodes, and traffic between the
two nodes can be mapped onto the LSP tunnels according to local
policy. Although traffic engineering (that is, performance
optimization of operational networks) is expected to be an important
application of this specification, the extended RSVP protocol can be
used in a much wider context.
The purpose of this document is to describe the use of RSVP to
establish LSP tunnels. The intent is to fully describe all the
objects, packet formats, and procedures required to realize
interoperable implementations. A few new objects are also defined
that enhance management and diagnostics of LSP tunnels.
The document also describes a means of rapid node failure detection
via a new HELLO message.
All objects and messages described in this specification are optional
with respect to RSVP. This document discusses what happens when an
object described here is not supported by a node.
Throughout this document, the discussion will be restricted to
unicast label switched paths. Multicast LSPs are left for further
Hosts and routers that support both RSVP  and Multi-Protocol Label
Switching  can associate labels with RSVP flows. When MPLS and
RSVP are combined, the definition of a flow can be made more
flexible. Once a label switched path (LSP) is established, the
traffic through the path is defined by the label applied at the
ingress node of the LSP. The mapping of label to traffic can be
accomplished using a number of different criteria. The set of
packets that are assigned the same label value by a specific node are
said to belong to the same forwarding equivalence class (FEC) (see
), and effectively define the "RSVP flow." When traffic is mapped
onto a label-switched path in this way, we call the LSP an "LSP
Tunnel". When labels are associated with traffic flows, it becomes
possible for a router to identify the appropriate reservation state
for a packet based on the packet's label value.
The signaling protocol model uses downstream-on-demand label
distribution. A request to bind labels to a specific LSP tunnel is
initiated by an ingress node through the RSVP Path message. For this
purpose, the RSVP Path message is augmented with a LABEL_REQUEST
object. Labels are allocated downstream and distributed (propagated
upstream) by means of the RSVP Resv message. For this purpose, the
RSVP Resv message is extended with a special LABEL object. The
procedures for label allocation, distribution, binding, and stacking
are described in subsequent sections of this document.
The signaling protocol model also supports explicit routing
capability. This is accomplished by incorporating a simple
EXPLICIT_ROUTE object into RSVP Path messages. The EXPLICIT_ROUTE
object encapsulates a concatenation of hops which constitutes the
explicitly routed path. Using this object, the paths taken by
label-switched RSVP-MPLS flows can be pre-determined, independent of
conventional IP routing. The explicitly routed path can be
administratively specified, or automatically computed by a suitable
entity based on QoS and policy requirements, taking into
consideration the prevailing network state. In general, path
computation can be control-driven or data-driven. The mechanisms,
processes, and algorithms used to compute explicitly routed paths are
beyond the scope of this specification.
One useful application of explicit routing is traffic engineering.
Using explicitly routed LSPs, a node at the ingress edge of an MPLS
domain can control the path through which traffic traverses from
itself, through the MPLS network, to an egress node. Explicit
routing can be used to optimize the utilization of network resources
and enhance traffic oriented performance characteristics.
The concept of explicitly routed label switched paths can be
generalized through the notion of abstract nodes. An abstract node
is a group of nodes whose internal topology is opaque to the ingress
node of the LSP. An abstract node is said to be simple if it
contains only one physical node. Using this concept of abstraction,
an explicitly routed LSP can be specified as a sequence of IP
prefixes or a sequence of Autonomous Systems.
The signaling protocol model supports the specification of an
explicit path as a sequence of strict and loose routes. The
combination of abstract nodes, and strict and loose routes
significantly enhances the flexibility of path definitions.
An advantage of using RSVP to establish LSP tunnels is that it
enables the allocation of resources along the path. For example,
bandwidth can be allocated to an LSP tunnel using standard RSVP
reservations and Integrated Services service classes .
While resource reservations are useful, they are not mandatory.
Indeed, an LSP can be instantiated without any resource reservations
whatsoever. Such LSPs without resource reservations can be used, for
example, to carry best effort traffic. They can also be used in many
other contexts, including implementation of fall-back and recovery
policies under fault conditions, and so forth.
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 .
The reader is assumed to be familiar with the terminology in , 
A group of nodes whose internal topology is opaque to the ingress
node of the LSP. An abstract node is said to be simple if it
contains only one physical node.
Explicitly Routed LSP
An LSP whose path is established by a means other than normal IP
Label Switched Path
The path created by the concatenation of one or more label
switched hops, allowing a packet to be forwarded by swapping
labels from an MPLS node to another MPLS node. For a more precise
definition see .
A Label Switched Path
An LSP which is used to tunnel below normal IP routing and/or
Traffic Engineered Tunnel (TE Tunnel)
A set of one or more LSP Tunnels which carries a traffic trunk.
A set of flows aggregated by their service class and then placed
on an LSP or set of LSPs called a traffic engineered tunnel. For
further discussion see .
2.1. LSP Tunnels and Traffic Engineered Tunnels
According to , "RSVP defines a 'session' to be a data flow with a
particular destination and transport-layer protocol." However, when
RSVP and MPLS are combined, a flow or session can be defined with
greater flexibility and generality. The ingress node of an LSP can
use a variety of means to determine which packets are assigned a
particular label. Once a label is assigned to a set of packets, the
label effectively defines the "flow" through the LSP. We refer to
such an LSP as an "LSP tunnel" because the traffic through it is
opaque to intermediate nodes along the label switched path.
New RSVP SESSION, SENDER_TEMPLATE, and FILTER_SPEC objects, called
LSP_TUNNEL_IPv4 and LSP_TUNNEL_IPv6 have been defined to support the
LSP tunnel feature. The semantics of these objects, from the
perspective of a node along the label switched path, is that traffic
belonging to the LSP tunnel is identified solely on the basis of
packets arriving from the PHOP or "previous hop" (see ) with the
particular label value(s) assigned by this node to upstream senders
to the session. In fact, the IPv4(v6) that appears in the object
name only denotes that the destination address is an IPv4(v6)
address. When we refer to these objects generically, we use the
In some applications it is useful to associate sets of LSP tunnels.
This can be useful during reroute operations or to spread a traffic
trunk over multiple paths. In the traffic engineering application
such sets are called traffic engineered tunnels (TE tunnels). To
enable the identification and association of such LSP tunnels, two
identifiers are carried. A tunnel ID is part of the SESSION object.
The SESSION object uniquely defines a traffic engineered tunnel. The
SENDER_TEMPLATE and FILTER_SPEC objects carry an LSP ID. The
SENDER_TEMPLATE (or FILTER_SPEC) object together with the SESSION
object uniquely identifies an LSP tunnel
2.2. Operation of LSP Tunnels
This section summarizes some of the features supported by RSVP as
extended by this document related to the operation of LSP tunnels.
These include: (1) the capability to establish LSP tunnels with or
without QoS requirements, (2) the capability to dynamically reroute
an established LSP tunnel, (3) the capability to observe the actual
route traversed by an established LSP tunnel, (4) the capability to
identify and diagnose LSP tunnels, (5) the capability to preempt an
established LSP tunnel under administrative policy control, and (6)
the capability to perform downstream-on-demand label allocation,
distribution, and binding. In the following paragraphs, these
features are briefly described. More detailed descriptions can be
found in subsequent sections of this document.
To create an LSP tunnel, the first MPLS node on the path -- that is,
the sender node with respect to the path -- creates an RSVP Path
message with a session type of LSP_TUNNEL_IPv4 or LSP_TUNNEL_IPv6 and
inserts a LABEL_REQUEST object into the Path message. The
LABEL_REQUEST object indicates that a label binding for this path is
requested and also provides an indication of the network layer
protocol that is to be carried over this path. The reason for this
is that the network layer protocol sent down an LSP cannot be assumed
to be IP and cannot be deduced from the L2 header, which simply
identifies the higher layer protocol as MPLS.
If the sender node has knowledge of a route that has high likelihood
of meeting the tunnel's QoS requirements, or that makes efficient use
of network resources, or that satisfies some policy criteria, the
node can decide to use the route for some or all of its sessions. To
do this, the sender node adds an EXPLICIT_ROUTE object to the RSVP
Path message. The EXPLICIT_ROUTE object specifies the route as a
sequence of abstract nodes.
If, after a session has been successfully established, the sender
node discovers a better route, the sender can dynamically reroute the
session by simply changing the EXPLICIT_ROUTE object. If problems
are encountered with an EXPLICIT_ROUTE object, either because it
causes a routing loop or because some intermediate routers do not
support it, the sender node is notified.
By adding a RECORD_ROUTE object to the Path message, the sender node
can receive information about the actual route that the LSP tunnel
traverses. The sender node can also use this object to request
notification from the network concerning changes to the routing path.
The RECORD_ROUTE object is analogous to a path vector, and hence can
be used for loop detection.
Finally, a SESSION_ATTRIBUTE object can be added to Path messages to
aid in session identification and diagnostics. Additional control
information, such as setup and hold priorities, resource affinities
(see ), and local-protection, are also included in this object.
Routers along the path may use the setup and hold priorities along
with SENDER_TSPEC and any POLICY_DATA objects contained in Path
messages as input to policy control. For instance, in the traffic
engineering application, it is very useful to use the Path message as
a means of verifying that bandwidth exists at a particular priority
along an entire path before preempting any lower priority
reservations. If a Path message is allowed to progress when there
are insufficient resources, then there is a danger that lower
priority reservations downstream of this point will unnecessarily be
preempted in a futile attempt to service this request.
When the EXPLICIT_ROUTE object (ERO) is present, the Path message is
forwarded towards its destination along a path specified by the ERO.
Each node along the path records the ERO in its path state block.
Nodes may also modify the ERO before forwarding the Path message. In
this case the modified ERO SHOULD be stored in the path state block
in addition to the received ERO.
The LABEL_REQUEST object requests intermediate routers and receiver
nodes to provide a label binding for the session. If a node is
incapable of providing a label binding, it sends a PathErr message
with an "unknown object class" error. If the LABEL_REQUEST object is
not supported end to end, the sender node will be notified by the
first node which does not provide this support.
The destination node of a label-switched path responds to a
LABEL_REQUEST by including a LABEL object in its response RSVP Resv
message. The LABEL object is inserted in the filter spec list
immediately following the filter spec to which it pertains.
The Resv message is sent back upstream towards the sender, following
the path state created by the Path message, in reverse order. Note
that if the path state was created by use of an ERO, then the Resv
message will follow the reverse path of the ERO.
Each node that receives a Resv message containing a LABEL object uses
that label for outgoing traffic associated with this LSP tunnel. If
the node is not the sender, it allocates a new label and places that
label in the corresponding LABEL object of the Resv message which it
sends upstream to the PHOP. The label sent upstream in the LABEL
object is the label which this node will use to identify incoming
traffic associated with this LSP tunnel. This label also serves as
shorthand for the Filter Spec. The node can now update its "Incoming
Label Map" (ILM), which is used to map incoming labeled packets to a
"Next Hop Label Forwarding Entry" (NHLFE), see .
When the Resv message propagates upstream to the sender node, a
label-switched path is effectively established.
2.3. Service Classes
This document does not restrict the type of Integrated Service
requests for reservations. However, an implementation SHOULD support
the Controlled-Load service  and the Null Service .
2.4. Reservation Styles
The receiver node can select from among a set of possible reservation
styles for each session, and each RSVP session must have a particular
style. Senders have no influence on the choice of reservation style.
The receiver can choose different reservation styles for different
An RSVP session can result in one or more LSPs, depending on the
reservation style chosen.
Some reservation styles, such as FF, dedicate a particular
reservation to an individual sender node. Other reservation styles,
such as WF and SE, can share a reservation among several sender
nodes. The following sections discuss the different reservation
styles and their advantages and disadvantages. A more detailed
discussion of reservation styles can be found in .
2.4.1. Fixed Filter (FF) Style
The Fixed Filter (FF) reservation style creates a distinct
reservation for traffic from each sender that is not shared by other
senders. This style is common for applications in which traffic from
each sender is likely to be concurrent and independent. The total
amount of reserved bandwidth on a link for sessions using FF is the
sum of the reservations for the individual senders.
Because each sender has its own reservation, a unique label is
assigned to each sender. This can result in a point-to-point LSP
between every sender/receiver pair.
2.4.2. Wildcard Filter (WF) Style
With the Wildcard Filter (WF) reservation style, a single shared
reservation is used for all senders to a session. The total
reservation on a link remains the same regardless of the number of
A single multipoint-to-point label-switched-path is created for all
senders to the session. On links that senders to the session share,
a single label value is allocated to the session. If there is only
one sender, the LSP looks like a normal point-to-point connection.
When multiple senders are present, a multipoint-to-point LSP (a
reversed tree) is created.
This style is useful for applications in which not all senders send
traffic at the same time. A phone conference, for example, is an
application where not all speakers talk at the same time. If,
however, all senders send simultaneously, then there is no means of
getting the proper reservations made. Either the reserved bandwidth
on links close to the destination will be less than what is required
or then the reserved bandwidth on links close to some senders will be
greater than what is required. This restricts the applicability of
WF for traffic engineering purposes.
Furthermore, because of the merging rules of WF, EXPLICIT_ROUTE
objects cannot be used with WF reservations. As a result of this
issue and the lack of applicability to traffic engineering, use of WF
is not considered in this document.
2.4.3. Shared Explicit (SE) Style
The Shared Explicit (SE) style allows a receiver to explicitly
specify the senders to be included in a reservation. There is a
single reservation on a link for all the senders listed. Because
each sender is explicitly listed in the Resv message, different
labels may be assigned to different senders, thereby creating
SE style reservations can be provided using multipoint-to-point
label-switched-path or LSP per sender. Multipoint-to-point LSPs may
be used when path messages do not carry the EXPLICIT_ROUTE object, or
when Path messages have identical EXPLICIT_ROUTE objects. In either
of these cases a common label may be assigned.
Path messages from different senders can each carry their own ERO,
and the paths taken by the senders can converge and diverge at any
point in the network topology. When Path messages have differing
EXPLICIT_ROUTE objects, separate LSPs for each EXPLICIT_ROUTE object
must be established.
2.5. Rerouting Traffic Engineered Tunnels
One of the requirements for Traffic Engineering is the capability to
reroute an established TE tunnel under a number of conditions, based
on administrative policy. For example, in some contexts, an
administrative policy may dictate that a given TE tunnel is to be
rerouted when a more "optimal" route becomes available. Another
important context when TE tunnel reroute is usually required is upon
failure of a resource along the TE tunnel's established path. Under
some policies, it may also be necessary to return the TE tunnel to
its original path when the failed resource becomes re-activated.
In general, it is highly desirable not to disrupt traffic, or
adversely impact network operations while TE tunnel rerouting is in
progress. This adaptive and smooth rerouting requirement
necessitates establishing a new LSP tunnel and transferring traffic
from the old LSP tunnel onto it before tearing down the old LSP
tunnel. This concept is called "make-before-break." A problem can
arise because the old and new LSP tunnels might compete with each
other for resources on network segments which they have in common.
Depending on availability of resources, this competition can cause
Admission Control to prevent the new LSP tunnel from being
established. An advantage of using RSVP to establish LSP tunnels is
that it solves this problem very elegantly.
To support make-before-break in a smooth fashion, it is necessary
that on links that are common to the old and new LSPs, resources used
by the old LSP tunnel should not be released before traffic is
transitioned to the new LSP tunnel, and reservations should not be
counted twice because this might cause Admission Control to reject
the new LSP tunnel.
A similar situation can arise when one wants to increase the
bandwidth of a TE tunnel. The new reservation will be for the full
amount needed, but the actual allocation needed is only the delta
between the new and old bandwidth. If policy is being applied to
PATH messages by intermediate nodes, then a PATH message requesting
too much bandwidth will be rejected. In this situation simply
increasing the bandwidth request without changing the
SENDER_TEMPLATE, could result in a tunnel being torn down, depending
upon local policy.
The combination of the LSP_TUNNEL SESSION object and the SE
reservation style naturally accommodates smooth transitions in
bandwidth and routing. The idea is that the old and new LSP tunnels
share resources along links which they have in common. The
LSP_TUNNEL SESSION object is used to narrow the scope of the RSVP
session to the particular TE tunnel in question. To uniquely
identify a TE tunnel, we use the combination of the destination IP
address (an address of the node which is the egress of the tunnel), a
Tunnel ID, and the tunnel ingress node's IP address, which is placed
in the Extended Tunnel ID field.
During the reroute or bandwidth-increase operation, the tunnel
ingress needs to appear as two different senders to the RSVP session.
This is achieved by the inclusion of the "LSP ID", which is carried
in the SENDER_TEMPLATE and FILTER_SPEC objects. Since the semantics
of these objects are changed, a new C-Types are assigned.
To effect a reroute, the ingress node picks a new LSP ID and forms a
new SENDER_TEMPLATE. The ingress node then creates a new ERO to
define the new path. Thereafter the node sends a new Path Message
using the original SESSION object and the new SENDER_TEMPLATE and
ERO. It continues to use the old LSP and refresh the old Path
message. On links that are not held in common, the new Path message
is treated as a conventional new LSP tunnel setup. On links held in
common, the shared SESSION object and SE style allow the LSP to be
established sharing resources with the old LSP. Once the ingress
node receives a Resv message for the new LSP, it can transition
traffic to it and tear down the old LSP.
To effect a bandwidth-increase, a new Path Message with a new LSP_ID
can be used to attempt a larger bandwidth reservation while the
current LSP_ID continues to be refreshed to ensure that the
reservation is not lost if the larger reservation fails.
2.6. Path MTU
Standard RSVP  and Int-Serv  provide the RSVP sender with the
minimum MTU available between the sender and the receiver. This path
MTU identification capability is also provided for LSPs established
Path MTU information is carried, depending on which is present, in
the Integrated Services or Null Service objects. When using
Integrated Services objects, path MTU is provided based on the
procedures defined in . Path MTU identification when using Null
Service objects is defined in .
With standard RSVP, the path MTU information is used by the sender to
check which IP packets exceed the path MTU. For packets that exceed
the path MTU, the sender either fragments the packets or, when the IP
datagram has the "Don't Fragment" bit set, issues an ICMP destination
unreachable message. This path MTU related handling is also required
for LSPs established via RSVP.
The following algorithm applies to all unlabeled IP datagrams and to
any labeled packets which the node knows to be IP datagrams, to which
labels need to be added before forwarding. For labeled packets the
bottom of stack is found, the IP header examined.
Using the terminology defined in , an LSR MUST execute the
1. Let N be the number of bytes in the label stack (i.e, 4 times the
number of label stack entries) including labels to be added by
2. Let M be the smaller of the "Maximum Initially Labeled IP Datagram
Size" or of (Path MTU - N).
When the size of an IPv4 datagram (without labels) exceeds the value
If the DF bit is not set in the IPv4 header, then
(a) the datagram MUST be broken into fragments, each of whose
size is no greater than M, and
(b) each fragment MUST be labeled and then forwarded.
If the DF bit is set in the IPv4 header, then
(a) the datagram MUST NOT be forwarded
(b) Create an ICMP Destination Unreachable Message:
i. set its Code field  to "Fragmentation Required and
ii. set its Next-Hop MTU field  to M
(c) If possible, transmit the ICMP Destination Unreachable
Message to the source of the of the discarded datagram.
When the size of an IPv6 datagram (without labels) exceeds the
value of M,
(a) the datagram MUST NOT be forwarded
(b) Create an ICMP Packet too Big Message with the Next-Hop
link MTU field  set to M
(c) If possible, transmit the ICMP Packet too Big Message to
the source of the of the discarded datagram.
3. LSP Tunnel related Message Formats
Five new objects are defined in this section:
Object name Applicable RSVP messages
RECORD_ROUTE Path, Resv
New C-Types are also assigned for the SESSION, SENDER_TEMPLATE, and
Detailed descriptions of the new objects are given in later sections.
All new objects are OPTIONAL with respect to RSVP. An implementation
can choose to support a subset of objects. However, the
LABEL_REQUEST and LABEL objects are mandatory with respect to this
The LABEL and RECORD_ROUTE objects, are sender specific. In Resv
messages they MUST appear after the associated FILTER_SPEC and prior
to any subsequent FILTER_SPEC.
The relative placement of EXPLICIT_ROUTE, LABEL_REQUEST, and
SESSION_ATTRIBUTE objects is simply a recommendation. The ordering
of these objects is not important, so an implementation MUST be
prepared to accept objects in any order.
3.1. Path Message
The format of the Path message is as follows:
<Path Message> ::= <Common Header> [ <INTEGRITY> ]
[ <EXPLICIT_ROUTE> ]
[ <SESSION_ATTRIBUTE> ]