Internet Engineering Task Force Dimitry Haskin
Internet Draft Ram Krishnan
Expires: May 2001 Axiowave Networks
November 2000
A Method for Setting an Alternative Label Switched Paths
to Handle Fast Reroute
draft-haskin-mpls-fast-reroute-05.txt
Status
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 describes a method for setting up an alternative label
switched path to handle fast reroute of traffic upon a failure in a
primary label switched path in Multi-protocol Label Switching (MPLS)
network.
Haskin, et al. [Page 1]
Internet Draft draft-haskin-mpls-fast-reroute-05.txt November 2000
Table of Contents
1. Introduction.....................................................2
2. Alternative Path Arrangement.....................................3
3. 1:1 protection...................................................6
4. 1:N protection...................................................6
5. Restoration Shortcuts............................................7
6. Elementary link level protection scheme..........................8
7. Bandwidth Reservation Considerations.............................8
8. Intellectual Property Considerations.............................9
9. Acknowledgments..................................................9
10. References.......................................................9
11. Authors' Addresses...............................................9
1. Introduction
The ability to quickly reroute traffic around a failure or
congestion in a label switched path (LSP) can be important in
mission critical MPLS networks. When an established label switched
path becomes unusable (e.g. due to a physical link or switch
failure) data may need to be re-routed over an alternative path.
Such an alternative path can be established after a primary path
failure is detected or, alternatively, it can be established
beforehand in order to reduce the path switchover time.
Pre-established alternative paths are essential where packet loss
due to an LSP failure is undesirable. Since it may take a
significant time for a device on a label switched path to detect a
distant link failure, it may continue sending packets along the
primary path. As soon as such packets reach a switch that is aware
of the failure, packets must be immediately rerouted by the switch
to an alternative path away from the failure if loss of data is to
be avoided. Since it is impossible to predict where failure may
occur along an LSP tunnel, it might involve complex computations and
extensive signaling to establish alternative paths to protect the
entire tunnel. In the extreme, to fully protect an LSP tunnel,
alternative paths might be established at each intermediate switch
along the primary LSP.
This document defines a method for setting alternative label
switched paths with the objective to provide a single failure
protection in such a manner that facilitates quick restoration
comparable to 50 milliseconds provided in SONET self-healing rings
and at the same time minimizes alternative path computation
complexity and signaling requirements. It also can provide in-band
means for quick detection of link and switch failures or congestion
along a primary path without resorting to an out of band signaling
mechanism. Both one-to-one (1:1) protection and many-to-one (1:N)
protection can be achieved with the proposed approach as described
in this document.
Haskin, et al. Expires May 2001 [Page 2]
Internet Draft draft-haskin-mpls-fast-reroute-05.txt November 2000
In order for the presented method to work, it is important that
network topology and policy allow the establishment of a backup LSP
between the endpoint switches of the protected LSP tunnel such that,
with the exception of the tunnel endpoint switches, the backup LSP
does not share any resources with the path that it intends to
protect.
The fast reroute support can be facilitated with additional
extensions incorporated in the MPLS signaling protocols such as RSVP
or CR-LDP. These extensions are not defined in this document.
2. Alternative Path Arrangement
The main idea behind the presented method is to reverse traffic at
the point of failure of the protected LSP back to the source switch
of the protected LSP such that the traffic flow can be then
redirected via a parallel LSP between source and destination
switches of the protected LSP tunnel.
Referring to Figure 1, there is an MPLS network consisting of 7
interconnected switches.
Figure 1:
+--------+ 24 +--------+ 46 +--------+
+-->| Switch |------->| Switch |------->| Switch |---+
: | 2 |--------| 4 |--------| 6 | :
: | | | | | | :
12 : +--------+ +--------+ +--------+ : 67
: / / / \ :
: / / / \ V
+--------+ 31 +--------+ 53 +--------+ 75 +--------+
| Switch |<-------| Switch |<-------| Switch |<......| Switch |
| 1 |--------| 3 |--------| 5 |-------| 7 |
=>| |=======>| |=======>| |======>| |=>
+--------+ 13 +--------+ 35 +--------+ 57 +--------+
The following terminology is used for purpose of describing the
method:
A portion of a label switched path that is to be protected by an
alternative path is referred as 'protected path segment'. Only
failures within the protected path segment, which may include the
entire primary path, are subject to fast reroute to the alternative
path. A primary LSP between switches 1 and 7 is shown by a double-
dashed links labeled 13, 35, and 57. Arrows indicate direction of
the data traffic.
The switch at the ingress endpoint of the protected path segment is
referred as 'the source switch'. Switch 1 in Figure 1 is the source
switch in our example of a protected path.
Haskin, et al. Expires May 2001 [Page 3]
Internet Draft draft-haskin-mpls-fast-reroute-05.txt November 2000
The switch at the egress endpoint of the protected path segment is
referred as 'the destination switch'. Switch 7 in Figure 1 is the
destination switch in our example of a protected path.
The switches between the source switch and the destination switch
along the protected path are referred as protected switches.
The switch immediately preceding the destination switch along the
protected path segment is referred as the last hop switch. Switch 5
in Figure 1 is the last hop switch for the protected path.
The essence of the presented method is that an alternative
unidirectional label switched path is established in the following
way:
The initial segment of the alternative LSP runs between the last
hop switch and the source switch in the reverse direction of the
protected path traversing through every protected switch between
the last hop switch and the source switch. The dashed line between
switches 5 and 1 illustrates such a segment of the alternative
path. Alternatively, the initial LSP segment can be set from the
destination switch to the source switch in the reverse direction
of the protected path traversing through every protected switch
between the destination switch and the source switch. The dashed
line between switches 7 and 1 illustrates the initial path segment
that is set in this way.
The second and final segment of the alternative path is set
between the source switch and the destination switch along a
transmission path that does not utilize any protected switches. It
is not an intention of this document to specify procedures for
calculating such a path. The dashed line between Switches 1 and 7
through Switches 2, 4, and 6 illustrates the final segment of the
alternative path.
The initial and final segments of the alternative path are linked to
form an entire alternative path from the last hop switch to the
destination switch. In Figure 1 the entire alternative path consists
of the LSP links labeled 53, 31, 12, 24, 46, and 67 if the
alternative path originates at the last hop switch. Alternatively,
the entire alternative path consists of the LSP links labeled 75,
53, 31, 12, 24, 46, and 67 if the alternative path originates at the
destination switch of the primary path.
As soon as a failure along the protected path is detected, an
operational switch at the ingress of the failed link reroutes
incoming traffic around the failure or congestion by redirecting
this traffic into the alternative LSP traversing the switch in the
reverse direction of the primary LSP according to the procedures
described in the following sections of the document.
Haskin, et al. Expires May 2001 [Page 4]
Internet Draft draft-haskin-mpls-fast-reroute-05.txt November 2000
The presented method of setting the alternative label switched path
has the following benefits:
- Path computation complexity is greatly reduced. Only a single
additional path between the source and destination switches of
the protected path segment needs to be calculated. Moreover,
both primary and alternative path computations can be localized
at a single switch avoiding problems that can arise when
computations are distributed among multiple switches.
- The amount of LSP setup signaling is minimized. With small
extensions to RSVP or LDP (described in separated documents), a
single switch at ingress of the protected path can initiate
label allocations for both primary and alternative paths.
- Optionally, presence of traffic on the alternative path segment
that runs in the reverse direction of the primary path can be
used as an indication of a failure or congestion of a downstream
link along the primary path. As soon as the source switch
detects the reverse traffic flow, it may stop sending traffic
downstream of the primary path and start sending data traffic
directly along the final alternative path segment.
It is fair to note that this technique increases the likelihood
of data packet reordering during the path rerouting process.
Therefore benefits of the reducing the alternative path latency
should be weighed against possible problems associated with
short term packet reordering. On a positive side, if multiple
microflows are aggregated in a single protected LSP tunnel, only
a very limited number of microflows may be affected by such
packet reordering. Additionally, the impact of reordering on any
single microflow may be minimal.
The described in-band signaling of an LSP failure to the source
switch does not exclude other methods of propagating an error
condition back to the source.
It also can be noted that if the alternative label switched path is
originated at the destination switch of the primary path, it forms a
'loop-back' LSP that originates and terminates at this switch.
Therefore in this case it is possible to verify integrity of the
entire alternative path by simply sending a probe packet from the
destination switch along the alternative path and asserting that the
packet arrives back to the destination switch. When this technique
is used to assert the path integrity, the care must be taken that
the limited diagnostic traffic is not interpreted as an indication
of a primary path failure that triggers data rerouting at the source
switch.
Haskin, et al. Expires May 2001 [Page 5]
Internet Draft draft-haskin-mpls-fast-reroute-05.txt November 2000
3. 1:1 protection
If the 1:1 path protection is desired, an individual backup LSP is
set for each LSP that needs to be protected as described in section
2. When a switch detects that a downstream link has failed, it
simply splices the traffic onto the alternative LSP. Referring to
Figure 1, if the link between the Switch 3 and Switch 5 fails,
Switch 3 accomplishes the fast reroute by swapping the incoming MPLS
label 13 of the primary path with the outgoing MPLS label 31 of the
alternative path. In this example the primary and alternative paths
are linked at Switch 3 forming the following label switched path for
the traffic flow: 13->31->12->24->46->67.
4. 1:N protection
In the case of the 1:N protection a single alternative path can be
used for protection of more than one LSP between the same source and
destination switches. The difference in rerouting LSPs the 1:N
protection case is that, rather than splicing protected traffic into
the alternative LSP, it may be necessary to use the MPLS label
stacking to tunnel protected traffic via the backup LSP to the
destination switch as described below.
A switch detecting failure of a downstream link, first swaps the
incoming MPLS label of each protected LSP with the respective
incoming label that identifies that LSP at the destination switch
and then pushes the outgoing label of the backup LSP to the top of
the forwarded MPLS packets. In essence, the protected MPLS packets
are encapsulated inside of the backup LSP and emerge at the backup
tunnel tail at the egress switch with their respective labels known
to that switch.
Referring to Figure 1 and assuming that global label space is used
at the destination switch, if the link between the Switch 3 and
Switch 5 fails, Switch 3 swaps incoming MPLS label 13 of the
protected LSP with label 57 (incoming label at Switch 7) and then
encapsulates the resulting packet into the backup tunnel by pushing
label 31 to the top of the forwarded MPLS packets.
Needless to say in order for this scheme to work, each router in the
protected path must be aware what labels are used at the egress LSR
for each protected LSP. Such knowledge can be propagated with the
appropriate extensions incorporated into signaling protocols such as
RSVP or CR-LDP.
A single segment of a tunnel between source and destination switches
can be used to protect multiple LSP segments that originate and
terminate on these switches as long as this segment of the backup
tunnel is completely disjoint from each protected LSP segment except
for the source and destination switches. In such a case the reverse
segments of backup path merge into the disjoint segment of the
backup path at the source switch of the protected LSPs as
Haskin, et al. Expires May 2001 [Page 6]
Internet Draft draft-haskin-mpls-fast-reroute-05.txt November 2000
illustrated in Figure 2. In Figure 2, dashed lines represent
protected LSPs and double-dashed lines represent backup LSP tunnels.
Figure 2:
+--+ +---+ +--+
| |=======>|LSR|========>|D |
| | +---+ |E |
| | +---+ +---+ |S |
| S|<===| |<===| |<===|T |
| O|--->|LSR|---> LSR|--->|I |
| U| +---+ +---+ |N |
| R| |A |
| C| +---+ +---+ |T |
| E|<===| |<===| |<===|I |
| |--->|LSR|--->|LSR|--->|O |
| | +---+ +---+ |N |
+--+ +--+
5. Restoration Shortcuts
Some types of applications require bounded end-to-end transmission
delays to deliver useful services. A notable example is the Voice
over IP (VoIP) service which requires end-to-end delays that do not
exceed 400 ms for an acceptable level of service. VoIP is also a
prime candidate for the fast reroute services. Since most of the
voice codecs in use today operate in the range of 20-50 ms latency,
the network component is left with around 300 ms of the end-to-end
delay limit.
Given the above considerations, it is important that, when
restoration provisions are made for a delay sensitive service,
transmission delays over an alternative path would not exceed an
acceptable limit. Since a number of the current network providers
are capable to guarantee network transport delay that do not exceed
80 ms on their backbone, it appears that in some cases it will be
possible to use the proposed restoration technique with a single
alternative path. It allows for at most 200 ms round trip delay over
a reverse path segment plus at most 100 ms delay over a disjoint
backup path segment. However in other cases it may be necessary to
introduce restoration shortcuts as described below to satisfy the
VoIP latency requirement during restoration.
Restoration shortcuts are achieved by allowing selected transit
routers in the primary LSP to establish one or more 'shortcut'
alternative LSPs to the egress router as illustrated in Figure 3. In
this illustration, primary link failures that may occur downstream
of LSR B are rerouted over the shortcut LSP from LSR B to the
destination of LSP being backed up. In illustrated example the
shortcut LSP merges into the backup LSP at LSR D.
Haskin, et al. Expires May 2001 [Page 7]
Internet Draft draft-haskin-mpls-fast-reroute-05.txt November 2000
Figure 3:
+--+ +---+ +--+
| |------------>|LSR|------------>|D |
| | | D | |E |
| S| +---+ |S |
| O| ^ |T |
| U| | |I |
| R| | |N |
| C| +---+ +---+ +---+ |A |
| E|<---|LSR|<---|LSR|<---|LSR|<---|T |
| |===>| A |===>| B |===>| C |===>|I |
| | +---+ +---+ +---+ |O |
| | |N |
+--+ +--+
6. Elementary link level protection scheme
If only link-level protection is desired, an alternative path
between link endpoints can be set up to protect each link. Such a
scheme can be viewed as a degenerate case of this proposal in which
the link endpoints constitute the source and destination endpoints
in the described approach.
7. Bandwidth Reservation Considerations
Generally there is no need to exclusively allocate bandwidth
resources to the alternate LSP. The holding priority of the primary
LSP can be used as traffic-triggered resource preemption priority
for the alternate LSP in case the primary LSP fails and traffic is
switched to the alternate LSP as described in this document. What we
call here the traffic-triggered priority is the preemption priority
assigned to an LSP that is utilized only when there is traffic
present on that LSP. When there is no traffic, other LSPs sharing
the interface should get full access to bandwidth and other system
resources. Consequently, if the traffic-triggered priority of the
alternative LSP is greater than the holding priorities of the other
LSPs using an interface in the alternate path, the alternate LSP can
preempt bandwidth and other system resources as soon as traffic gets
rerouted via the alternate LSP. This enables high-priority LSPs,
which are being rerouted, to preempt resources from lower priority
LSPs without explicit bandwidth reservation for the alternate path.
Of course, if bandwidth efficiency is not an issue, bandwidth
resources can be explicitly reserved for the alternate LSP also.
An extension to existing signaling protocols such as RSVP and LDP
may be needed to indicate that traffic-triggered resource preemption
is requested for a particular LSP as opposed to the setup priority
preemption.
Haskin, et al. Expires May 2001 [Page 8]
Internet Draft draft-haskin-mpls-fast-reroute-05.txt November 2000
8. Intellectual Property Considerations
IETF has been informed of possible intellectual property protection
for some or all of the technologies disclosed in this document.
9. Acknowledgments
This document has benefited from discussions with Jim Boyle, Robert
Boyd, and Alan Hannan. We also thank Ken Schroder, Jeff Parker and
Yanhe Fan for their comments on the document.
10. References
[1] Rosen, E. et al., "Multiprotocol Label Switching Architecture",
Internet Draft, draft-ietf-mpls-arch-07.txt, July 2000.
[2] Awduche, D. et al., "Requirements for Traffic Engineering over
MPLS", RFC-2702.
11. Authors' Addresses
Dimitry Haskin
Axiowave Networks, Inc.
100 Nickerson Road
Marlborough, MA 01752
E-mail: dhaskin@axiowave.com
Ram Krishnan
Axiowave Networks, Inc.
100 Nickerson Road
Marlborough, MA 01752
E-mail: ram@axiowave.com
Haskin, et al. Expires May 2001 [Page 9]