Networking Working Group K. Shiomoto
Internet-Draft NTT
Intended Status: Informational
A. Farrel
Expires: May 9, 2011 Huawei Technologies
November 9, 2010
Advice on When It is Safe to Start Sending Data on
Label Switched Paths Established Using RSVP-TE
draft-shiomoto-ccamp-switch-programming-03.txt
Abstract
The Resource Reservation Protocol (RSVP) has been extended to support
Traffic Engineering (TE) in Multiprotocol Label Switching (MPLS) and
Generalized MPLS (GMPLS) networks. The protocol enables signaling
exchanges to establish Label Switched Paths (LSPs) that traverse
nodes and links to provide end-to-end data paths. Each node is
programmed with "cross-connect" information as the signaling messages
are processed. The cross-connection information instructs the node
how to forward data that it receives.
End points of an LSP need to know when it is safe to start sending
data so that it is not misdelivered, and so that safety issues
specific to the data plane technology are satiefied. Likewise, all
label switching routers along the path of the LSP need to know when
to programme their data planes relative to sending control plane
messages.
This document clarifies and summarises the RSVP-TE protocol exchanges
with relation to the programming of cross-connects along an LSP for
both unidirectional and bidirectional LSPs. This document does not
define any new procedures or protocol extensions, and defers
completely to the documents that provide normative references. The
clarifications set out in this document may also be used to help
interpret LSP establishment performance figures for MPLS-TE and GMPLS
devices.
Status of this Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
Shiomoto and Farrel [Page 1]
Internet-Draft When to Start Sending Data November 2010
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."
This Internet-Draft will expire on January 7, 2011.
Copyright Notice
Copyright (c) 2010 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
1. Introduction
The Resource Reservation Protocol (RSVP) [RFC2205] has been extended
to support Traffic Engineering (TE) in Multiprotocol Label Switching
(MPLS) and Generalized MPLS (GMPLS) networks [RFC3209], [RFC3473].
The protocol enables signaling exchanges to establish Label Switched
Paths (LSPs) that traverse nodes and links to provide end-to-end data
paths. Each node is programmed with "cross-connect" information as
the signaling messages are processed. The cross-connection
information instructs the node how to forward data that it receives.
In some technologies this requires configuration of physical devices,
while in others it may involve the exchange of commands between
different components of the node. The nature of a cross-connect is
described further in Section 1.1.1.
End points of an LSP need to know when it is safe to start sending
data. In this context "safe" has two meanings. The first issue is
that the sender needs to know that the data path has been fully
established, setting up the cross-connects and removing any old,
incorrect forwarding instructions, so that data will be delivered to
the intended destination. The other meaning of "safe" is that in
optical technologies lasers must not be turned on until the correct
cross-connects have been put in place to ensure that service
personnel are not put at risk.
Similarly, all Label Switching Routers (LSRs) along the path of the
LSP need to know when to programme their data planes relative to
Shiomoto and Farrel [Page 2]
Internet-Draft When to Start Sending Data November 2010
sending control plane messages.
This document clarifies and summarises the RSVP-TE protocol exchanges
with relation to the programming of cross-connects along an LSP for
both unidireciotnal and bidirectional LSPs. Bidirectional LSPs, it
should be noted, are supported only in GMPLS. This document does not
define any new procedures or protocol extensions, and defers
completely to the documents that provide normative references.
The clarifications set out in this document may also be used to help
interpret LSP establishment performance figures for MPLS-TE and GMPLS
devices. For example, the dynamic provisioning performance metrics
set out in [RFC5814] need to be understood in the context of LSP
setup times and not in terms of control message exchange times that
are actually only a component of the whole LSP establishment process.
Note that it would be really useful for implementations if this
document could definitively identify when it is safe for any LSR to
forward the Path or Resv message before programming its cross-
connect, exploiting pipelining to try to minimise the total time to
set up the LSP. However, while this document gives advice and
identifies the issues to be considered, it is not possible to make
definitive statements about how much pipeling is safe since a node
can "know" much without first probing the network (for example, with
protocol extensions) and that would defeat the point of pipelining.
There are so many variables introduced by path length and by the
behavior of other nodes that an ingress might be limited to a very
pessimistic view for safety. Furthermore, it seems unlikely that an
implementation would necessarily give a full and frank description of
how long it takes to program and stabilize its cross-connects.
Nevertheless, this document identifies the issues and oportunties for
for pipelining in GMPLS systems.
1.1. Terminology
It is assumed that the reader is familiar with the basic message
flows of RSVP-TE as used in MPLS-TE and GMPLS. Refer to [RFC2205],
[RFC3209], [RFC3471], and [RFC3473] for more details.
1.1.1. What is a Cross-Connect?
In the context of this document, the concept of a "cross-connection"
should be taken to imply the data forwarding instructions installed
(that is, "programmed") at a network node (or "switch").
In packet MPLS networks, this is often refered to as the Incoming
Label Map (ILM) and Next Hop Label Forwarding Entry (NHLFE) [RFC3031]
which are sometimes considered together as entries in the Label
Shiomoto and Farrel [Page 3]
Internet-Draft When to Start Sending Data November 2010
Forwarding Information Base (LFIB) [RFC4221]. Where there is
admission control and resource reservation associated with the data
forwarding path (such as the allocation of data buffers) [RFC3209]
this can be treated as part of the cross-connect programming process
since before the resources are correctly allocated and reserved, the
LSP will not be available to forward data in the manner agreed to
during the signaling protocol exchange.
In non-packet networks (such as time-division multiplexing, or
optical switching networks) the cross-connect concept may be an
electronic cross-connect array or a transparent optical device (such
as a MEMS). In all cases, however, the concept applies to the
instructions that are programmed into the forwarding plane (that is,
the data plane) so that incoming data for the LSP on one port can be
correctly handled and forwarded out of another port.
2. Unidirectional MPLS-TE LSPs
[RFC3209] describes the RSVP-TE signaling and processing for MPLS-TE
packet-based networks. LSPs in these networks are unidirectional by
definition (there are no bidirectional capabilities in [RFC3209]).
Section 4.1.1.1 of [RFC3209] describes the processing that a node
does before sending a Resv message to its upstream neighbor.
The node then sends the new LABEL object as part of the Resv
message to the previous hop. The node SHOULD be prepared to
forward packets carrying the assigned label prior to sending the
Resv message.
This means that the cross-connect should be in place to support
traffic that may arrive at the node before the node sends the Resv.
This is clearly advisable because the upstream LSRs might otherwise
complete their cross-connections more rapidly and encourage the
ingress to start transmitting data with the risk that the node that
sent the Resv "early" would be unable to forward the data it received
and would be forced to drop it, or might accidentally send it along
the wrong LSP because of stale cros-connect information.
The use of "SHOULD" [RFC2119] in this text indicates that an
implementation could be constructed that sends a Resv before it is
ready to receive and forward data. This might be done simply because
the internal construction of the node means that the control plane
components cannot easily tell when the cross-connection has been
installed. Alternatively it might arise because the implementation is
aware that it will be slow and does not wish to hold up the
establishment of the LSP. In this latter case, the implementation is
choosing to pipeline the cross-connect programming with the protocol
Shiomoto and Farrel [Page 4]
Internet-Draft When to Start Sending Data November 2010
exchange taking a gamble that there will be other upstream LSRs that
may also take some time to process, and it will in any case be some
time before the ingress actually starts to send data. It should be
noted, however, that as well as the risks described in the previous
paragraph, a node that behaves like this must include a mechanism to
report a failure to chase the Resv message (using a PathErr) in the
event that the pipelined cross-connect processing fails.
3. GMPLS LSPs
GMPLS [RFC3945] extends RSVP-TE signaling for use in networks of
different technonlogies [RFC3471], [RFC3473]. This means that RSVP-TE
signaling may be used in MPLS packet switching networks, as well as
layer two networks (Ethernet, Frame Relay, ATM), time-division
multiplexing networks (TDM, i.e., SONET and SDH), wavelength division
multiplexing networks (WDM), and fiber switched network.
The introdution of these other technologies, specifically the optical
technologies, brings about the second definition of the "safe"
commencement of data transmission as described in Seciotn 1. That is,
there is a physical safety issue that means that the lasers should
not be enabled until the corss-connects are correctly in place.
GMPLS supports unidirectional and bidirectional LSPs. These are split
into separate sections for discussion. The processing rules are
inherited from [RFC3209] unless they are specifially modified by
[RFC3471] and [RFC3473].
3.1. Unidirectional LSPs
Unidirectional LSP processing would be the same as that described in
Section 2 except for the use of the Suggested_Label object defined in
[RFC3473]. This object allows an upstream LSR to 'suggest' to its
downstream neighbor the label that should be used for forward-
direction data by including the object on a Path message. The purpose
of this object is to help the downstream LSR in its choice of label,
but it also makes it possible for the upstream LSR to 'pipeline'
programming its cross-connect with the RSVP-TE signaling exchanges.
That means that the cross-connect might be in place before the
signaling has completed (i.e., before a Resv message carrying a Label
object has been received at the upstream LSR).
We need to know when it is safe to start sending data. There are
three sources of information.
- Section 3.4 of [RFC3471] states:
... an ingress node should not transmit data traffic on a
Shiomoto and Farrel [Page 5]
Internet-Draft When to Start Sending Data November 2010
suggested label until the downstream node passes a label
upstream.
The implication here is that an ingress node may (safely) start to
transmit data when it receives a label in a Resv message.
- Section 2.5 of [RFC3473] states:
... an ingress node SHOULD NOT transmit data traffic using a
suggested label until the downstream node passes a corresponding
label upstream.
This is a confirmation of the first source.
- Section 4.1.1.1 of [RFC3209] states:
... The node then sends the new LABEL object as part of the Resv
message to the previous hop. The node SHOULD be prepared to
forward packets carrying the assigned label prior to sending the
Resv message.
In this text the word "prior" is very important. It means that the
cross-connect must be in place for forward traffic before the Resv
is sent. In other words, each of the the transit nodes and the
egress node must finish making their cross-connects before they
the Resv message to their upstream neighbors.
Thus, as in Section 2, we can deduce that the ingress must not start
to transmit traffic until it has both received a Resv and has
programmed its own cross-connect.
3.2. Bidirectional LSPs
A bidirectional LSP is established with one signaling exchange of a
Path message from ingress to egress, and a Resv from egress to
ingress. The LSP itself is comprised of two sets of forwarding state,
one providing a path from the ingress to the egress (the forwards
data path), and one from the egress to the ingress (the reverse
data path).
3.2.1. Forwards Direction Data
The processing for the forwards direction direction data path is
exactly as described for a unidirectional LSP in Section 3.1.
Shiomoto and Farrel [Page 6]
Internet-Draft When to Start Sending Data November 2010
3.2.2. Reverse Direction Data
For the reverse direction data flow an Upstream_Label object is
carried in the Path message from each LSR to its downstream neighbor.
The Upstream_Label object tells the downstream LSR which label to use
for data being sent to the upstream LSR (that is, reverse direction
data). The use of the label is confirmed by the downstream LSR when
it sends a Resv message. Note that there is no explicit confirmation
of the label in the Resv message, but if the label was not acceptable
to the downstream LSR, it would return a PathErr message instead.
The upstream LSR must decide when to send the Path message relative
to when it programmes its cross-connect. That is, should it programme
the cross-connect before it sends the Path message, can it overlap
the programming with the exchange of messages, or must it wait until
it receives a Resv from its downstream neighbor?
The defining reference is Section 3.1 of [RFC3473]:
The Upstream_Label object MUST indicate a label that is valid for
forwarding at the time the Path message is sent.
In this text "valid for forwarding" should be taken to mean that it
is safe for the LSR that sends the Path message to receive data, and
that the LSR will forward data correctly. The text does not mean that
the label is "acceptable for use" (i.e., the label is available to be
cross-connected).
This point is clarified later in Section 3.1 of [RFC3473]:
Terminator nodes process Path messages as usual, with the exception
that the upstream label can immediately be used to transport data
traffic associated with the LSP upstream towards the initiator.
This is a clear statement that when a Path message has been fully
processed by an egress node, it is completely safe to transmit data
toward the ingress (i.e., reverse direction data).
From this we can deduce several things:
- An LSR must not wait to receive a Resv message before it programmes
the cross-connect for the reverse direction data. It must be ready
to receive data from the moment that the egress completes
processing the Path message that it receives (i.e., before it sends
a Resv back upstream).
- An LSR may expect to start receiving reverse direction data as soon
as it sends a Path message for a bidirectional LSP.
Shiomoto and Farrel [Page 7]
Internet-Draft When to Start Sending Data November 2010
- An LSR may make some assumptions about the time lag between sending
a Path message and the message reaching and being processed by the
egress. It may take advantage of this time lag to pipeline
programming the cross-connect.
3.3. ResvConf Message
The ResvConf message is used in standard RSVP [RFC2205] to let the
ingress confirm to the egress that the Resv has been succesfully
received, and what bandwidth has been reserved. In RSVP-TE [RFC3209]
and GMPLS [RFC3473], it is not expected that bandwidth will be
modified along the path of the LSP, so the purpose of the ResvConf
is reduced to a confirmation that the LSP has been successfully
established.
The egress may request that a ResvConf is sent by including a
Resv_Confirm object in the Resv message that it sends. When the
ingress receives the Resv message and sees the Resv_Confirm object,
it can respond with a ResvConf message.
It should be clear that this mechanism might provide a doubly secure
way for the egress to ensure that the reverse direction data path is
safely in place before transmitting data. That is, if the egress
waits until it receives a ResvConf message, it can be sure that the
whole LSP is in place.
However, this mechanism is excessive given the definitions presented
in Section 3.2.2, and would delay LSP setup by one end-to-end message
propagation cycle. It should be noted as well that the generation and
of the ResvConf message is not guaranteed. Furthermore, many (if not
most) GMPLS implementations neither request nor send ResvConf
messages. Therefore, an egress is not recommended to rely on the
receipt of a ResvConf as a way of knowing that it is safe to start
transmitting reverse direction data.
3.4. Administrative Status
GMPLS offers an additional tool for ensuring safety of the LSP. The
Administrative Status information is defined in Section 8 of
[RFC3471] and is carried in the Admin_Status Object defined in
Section 7. of [RFC3473].
This object allows an ingress to set up an LSP in "Administratively
Down" state. This state means ([RFC3471] that:
... the local actions related to the "administratively down" state
should be taken.
Shiomoto and Farrel [Page 8]
Internet-Draft When to Start Sending Data November 2010
In this state it is assumed that the LSP exists (i.e., the cross-
connects are all in place), but no data is transmitted (i.e., in
optical systems, the lasers are off ).
Additionally, the Admin_Status object allows the LSP to be put into
"Testing" state. This state means ([RFC3471]) that:
... the local actions related to the "testing" mode should be
taken.
This state allows the connectivity of the LSP to be tested without
actually exchanging user data. For example, in an optical system it
would be possible to run a data continuity test (using some external
coordination of errors). In a packet network, a connection
verification exchange (such as the in-band Virtual Circuit
Connectivity Verification described in Section 5.1.1 of [RFC5085])
could be used. Once connectivity has been verified, the LSP could be
put into active mode and the exchange of user data could commence.
These processes may be considered particularly important in systems
where the control plane processors are physically distinct from the
data plane cross-connects (for example, where there is a
communication protocol operating between the control plane processor
and the data plane switch) in which case the successful completion of
control plane signaling cannot necessarily be taken as evidence of
correct data plane programming.
4. Implications for Performance Metrics
The ability of LSRs to handle and propagate control plane messages
and to programme cross-connects varies considerably from device to
device according to switching technology, control plane connectivity,
and implementation. These factors influence how quickly an LSP can be
established.
Different applications have different requirements for the speed of
setup of LSPs, and this may be particularly important in recovery
scenarios. It is important for service providers considering the
deployment of MPLS-TE or GMPLS equipment to have a good benchmark for
the performance of the equipment. Similarly, it is important for
equipment vendors to be compared on a level playing field.
In order to provide a basis for comparison, [RFC5814] defines a
series of performance metrics to evaluate dynamic LSP provisioning
performance in MPLS-TE/GMPLS networks. Any use of such metrics must
be careful to understand what is being measured bearing in mind that
it is not enough to know that the control plane message has been
processed and forwarded: the cross-connect must be put in place
Shiomoto and Farrel [Page 9]
Internet-Draft When to Start Sending Data November 2010
before the LSP can be used. Thus, care must be taken to ensure that
devices are correctly conforming to the procedures clarified in
Section 2 of this document, and not simply forwarding control plane
messages with the intent to programme the cross-connects in the
background.
5. IANA Considerations
This document makes no requests for IANA action.
6. Security Considerations
This document does not define any network behavior and does not
introduce or seek to solve any security issues.
It may be noted that a clear understanding of when to start sending
data may reduce the risk of data being accidentally delivered to the
wrong place.
7. Acknowledgments
Thanks to Weiqiang Sun and Olufemi Komolafe for their review and
comments.
8. References
8.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC2205] Braden, R. (Ed.), Zhang, L., Berson, S., Herzog, S., and S.
Jamin, "Resource ReserVation Protocol -- Version 1
Functional Specification", RFC 2205, September 1997.
[RFC3209] Awduche, D., Berger, L., Gan, D., Li, T., Srinivasan, V.,
and G. Swallow, "RSVP-TE: Extensions to RSVP for LSP
Tunnels", RFC 3209, December 2001.
[RFC3471] Berger, L., Ed., "Generalized Multi-Protocol Label
Switching (GMPLS) Signaling Functional Description", RFC
3471, January 2003.
[RFC3473] Berger, L., "Generalized Multi-Protocol Label Switching
(GMPLS) Signaling Resource ReserVation Protocol-Traffic
Engineering (RSVP-TE) Extensions", RFC 3473, January 2003.
Shiomoto and Farrel [Page 10]
Internet-Draft When to Start Sending Data November 2010
[RFC3945] Mannie, E., Ed., "Generalized Multi-Protocol Label
Switching (GMPLS) Architecture", RFC 3945, October 2004.
8.2. Informative References
[RFC3031] Rosen, E., Viswanathan, A., and R. Callon, "Multiprotocol
Label Switching Architecture", RFC 3031, January 2001.
[RFC4221] Nadeau, T., Srinivasan, C., and Farrel, A., "Multiprotocol
Label Switching (MPLS) Management Overview", RFC 4221,
November 2005.
[RFC5085] Nadeau, T., Pignataro, C., "Pseudowire Virtual Circuit
Connectivity Verification (VCCV): A Control Channel for
Pseudowires", RFC 5085, December 2007.
[RFC5814] Sun, W., and Zhang, G., "Label Switched Path (LSP) Dynamic
Provisioning Performance Metrics in Generalized MPLS
Networks", RFC 5814, March 2010.
Authors' Addresses
Kohei Shiomoto
NTT Network Service Systems Laboratories
3-9-11 Midori
Musashino, Tokyo 180-8585
Japan
Phone: +81 422 59 4402
Email: shiomoto.kohei@lab.ntt.co.jp
Adrian Farrel
Huawei Technologies
Email: adrian.farrel@huawei.com
Shiomoto and Farrel [Page 11]
