Internet DRAFT - draft-litkowski-rtgwg-lfa-rsvpte-cooperation

draft-litkowski-rtgwg-lfa-rsvpte-cooperation






Routing Area Working Group                                  S. Litkowski
Internet-Draft                                               B. Decraene
Intended status: Standards Track                                  Orange
Expires: February 20, 2014                                   C. FilsFils
                                                                 K. Raza
                                                           Cisco Systems
                                                         August 19, 2013


                  Interactions between LFA and RSVP-TE
            draft-litkowski-rtgwg-lfa-rsvpte-cooperation-02

Abstract

   This document defines the behavior of a node supporting Loopfree
   Alternates (LFA) when the node has established RSVP TE tunnels.  It
   first describes the decisions to be made by the LFA mechanism with
   respect to the use of TE tunnels as LFA candidates.  Second, it
   discusses the use of RSVP TE tunnels as a way to complement the LFA
   coverage, illustrating how these technologies can benefit from each
   other.

Requirements Language

   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].

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
   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 February 20, 2014.

Copyright Notice





Litkowski, et al.       Expires February 20, 2014               [Page 1]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   Copyright (c) 2013 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.

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   2
   2.  LFA FRR and MPLS-TE interactions  . . . . . . . . . . . . . .   3
     2.1.  Use case : using MPLS LSP as LFA candidates . . . . . . .   3
     2.2.  Specifications of interactions between LFA and TE LSP . .   4
       2.2.1.  Having both a physical interface and a TE tunnel
               toward a LFA  . . . . . . . . . . . . . . . . . . . .   4
       2.2.2.  TE ingress LSP as LFA candidate . . . . . . . . . . .   4
       2.2.3.  Independence between LFA and TE FRR . . . . . . . . .   5
   3.  Operational considerations  . . . . . . . . . . . . . . . . .   7
     3.1.  Relevance of joint LFA FRR and RSVP-TE FRR deployments  .   7
     3.2.  Extending LFA coverage using RSVP-TE tunnels  . . . . . .   8
       3.2.1.  Creating multihop tunnel to extend topology . . . . .   8
       3.2.2.  Selecting multihop tunnels to extend topology . . . .   9
   4.  Security Considerations . . . . . . . . . . . . . . . . . . .  10
   5.  Contributors  . . . . . . . . . . . . . . . . . . . . . . . .  10
   6.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .  10
   7.  References  . . . . . . . . . . . . . . . . . . . . . . . . .  10
     7.1.  Normative References  . . . . . . . . . . . . . . . . . .  10
     7.2.  Informative References  . . . . . . . . . . . . . . . . .  10
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  11

1.  Introduction

   When a failure occurs in an IP network, the subsequent converge
   process often leads to traffic disruption.  Some mechanisms are
   available to limit traffic disruptions by pre-computing alternate
   paths and locally reroute over these as soon as the failure is
   detected.  Such techniques are commonly known as "protection
   mechanisms".  Currently, the protection mechanisms widely used in
   Service Provider networks are RSVP-TE Fast Reroute [RFC4090] and Loop
   Free Alternates [RFC5286].  RSVP-TE FRR permits full network coverage
   but with a quite high complexity in terms of operation, as well as
   potential scaling issues.  On the other hand, LFA offer a very easy,



Litkowski, et al.       Expires February 20, 2014               [Page 2]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   manageable, and scalable mechanism, but does not provide full
   coverage.

   This document discusses how LFA and RSVP-TE should interact.  It
   first describes how an LFA implementation should deal with existing
   RSVP TE tunnels established by the LFA node, as well as its behavior
   with respect to established IGP Shortcut tunnels [RFC3906].  Second,
   the document suggets the use of RSVP-TE tunnels to extend LFA
   coverage, and discusses the management and operational aspects of
   such a practice.

2.  LFA FRR and MPLS-TE interactions

   This section discusses the various interactions among LFA FRR and
   MPLS-TE FRR.  It starts with a simple example emphasizing the
   benefits of jointly using of LFA-FRR and MPLS-TE FRR, and then
   summarizes the requirements for the interactions between LFAs and
   MPLS-FRR.

2.1.  Use case : using MPLS LSP as LFA candidates

   In some cases, typically in ring shapped parts of network topologies,
   links cannot be protected by LFAs.  In the following topology, from
   the point of view of R5, LFAs are able to partiatlly protect (49% of
   the destination routers) from the failure of R3, while the failure of
   R4 is not covered at all.

   (30 routers) --- R1 ----(30)---- R2 --- (50 routers)
                    |               |
                   (5)             (30)
                    |               |
                    R3               R4 --- (20 routers)
                    |               |
                   (10)             (30)
                    |               |
                    ------- R5 ------

                         Figure 1


   Many networks deploy MPLS tunnels for traffic engineering and
   resiliency reasons.  To extend its benefit, an LFA implementation
   could take advantage of such existing MPLS tunnels.  In the exemple
   above, if R5 has established TE tunnels bypassing R4 and R3, these
   could be considerd as LFA candidates respectively protecting links
   from R5 to R4 and R3.





Litkowski, et al.       Expires February 20, 2014               [Page 3]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   In the following section, we provide a detailed summary of the
   behavior to be applied by an LFA implementation which would consider
   the existence of MPLS TE tunnels to improve its applicability.  The
   explicit configuration of such tunnels with the intent of improving
   LFA applicability is discussed in later sections.

2.2.  Specifications of interactions between LFA and TE LSP

   Here we summarize the normative requirements for the interaction
   between LFA FRR and MPLS TE tunnels.

2.2.1.  Having both a physical interface and a TE tunnel toward a LFA

   If a node S has both a physical interface and a TE tunnel to reach a
   LFA, it SHOULD use the physical interface unless :

   1.  The tunnel has been explicitly configured as an LFA candidate.

   2.  The tunnel does not pass through the link subject to LFA
       protection.

   In other words, if a node S has an IGP/LDP forwarding entry F1 with
   outgoing interface i1, and S originates a TE tunnel T2 terminating on
   direct neighbor N2 (for example : if a TE tunnel is provisionned for
   link protection), T2 has an outgoing interface i2 and N2 is best LFA
   for F1, then an implementation MUST NOT use T2 when programming LFA
   repair for F1 unless T2 is configured as an LFA candidate.

2.2.2.  TE ingress LSP as LFA candidate

   A TE LSP can be used as a virtual interface to reach a LFA if

   1.  The TE tunnel has been configured to allow its use as an LFA
       candidate.

   2.  The TE tunnel does not pass through the primary outgoing
       interface of D.

   This would permit to extend LFA coverage as described in
   [I-D.ietf-rtgwg-remote-lfa], in a controlled fashioned, as the
   tunnels used by the fast reroute mechanism are defined by
   configuration.

   In other words, if a node S has an IGP/LDP forwarding entry F1 with
   outgoing interface i1 and S originates a TE tunnel T1 terminating at
   node Y, then an implementation SHOULD support a local policy which
   instructs node S to consider Y as a virtual neighbor and hence
   include Y as part of the LFA FRR alternate computation.  In such



Litkowski, et al.       Expires February 20, 2014               [Page 4]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   case, an implementation MUST not use Y as an LFA for F1 if T1's
   outgoing interface is i1.

2.2.3.  Independence between LFA and TE FRR

2.2.3.1.  Tunnel head-end case

   Similar requirements can be expressed for TE IGP shortcut tunnels.

           -----C1 ---------
         /      |           \
        /       |            \
      PE1 ---- C2 --- C3 ---- PE2
                |    /
                 PE3

              Figure 2


   PE to Cx metrics are 50, Cx to Cx are 1


   A service provider is often providing traffic-engineered path for
   specific customer traffic (L3VPN, PW ...) to ensure path diversity or
   traffic constraints.  In the diagram above, we consider a TE tunnel
   T2 built on a non shortest path as follows : PE1->C2->C3->PE2 and IGP
   shortcut is activated on PE1 to make traffic to PE2 using T2.  Based
   on operational feedback, some implementations prevent LFA computation
   to run for an interface where a TE tunnel exists.  In our example, if
   LFA is activated on N, we would not be able to have a protection for
   PE3 destination as a tunnel exists on the interface.  This current
   observed behavior leads to a very limited coverage for LFA.  In the
   other hand, it is important to keep protection mechanisms independant
   as much as possible to keep implementation simple.  We propose the
   following approach :

   o  If an IP prefix is reachable through a TE tunnel, LFA must not
      compute a protection for it.

   o  If an IP prefix is reachable through a native IP path, LFA MUST
      compute a protection for it disregarding the presence of a tunnel
      or not on the primary interface.









Litkowski, et al.       Expires February 20, 2014               [Page 5]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   In other words, if a node S has an IGP/LDP forwarding entry F1 with
   outgoing interface i1 and an IGP/LDP forwarding entry F2 with
   outgoing interface onto a TE tunnel T2 (due to IGP shortcut
   [RFC3906]) and tunnel T2 has outgoing interface i2, then an
   implementation MUST support enabling LFA FRR for F1 and using TE FRR
   for F2 as long as i1 != i2.

   If i1 == i2, an implementation SHOULD allow for using LFA FRR backup
   for F1 and TE FRR backup for F2.

   The mechanisms for using TE tunnel as an LFA candidate, and RFC3906
   mechanisms MUST be de-correlated -- i.e an implementation MUST
   support TE tunnel configuration with RFC3906 only, as LFA candidate
   only, or both at the same time.

2.2.3.2.  Tunnel midpoint case

          -----C1 ---------
         /      |           \
        /       |            \
      PE1 ---- C2 --- C3 ---- PE2
                |    /
                 PE3

               Figure 3

   PE to Cx metrics are 50, except PE3-C3 (60), Cx to Cx are 1


   In the diagram above, we consider a TE tunnel T2 built on a non
   shortest path as follows : PE1->C2->C3->PE2 and IGP shortcut is
   activated on PE1 to make traffic to PE2 using T2.  C2 is a TE tunnel
   midpoint router.  In terms of forwarding, C2 has a MPLS TE forwarding
   entry for T2, as well as an IP forwarding entry to PE2.  As explained
   in previous sections, it would be too restrictive and would limit LFA
   benefit on C2 if C2 would not be able to compute an LFA for the IP
   forwarding entry to PE2 due to the presence of a transit tunnel.

   We propose the following approach for a midpoint router of a TE
   tunnel :

   o  MPLS TE forwarding entries MUST not be protected by LFA (if an
      operator wants protection, TE FRR could be enabled).

   o  IP forwarding entries MUST be protected by LFA disregarding the
      presence of a TE tunnel transiting through the primary interface
      of the destination.




Litkowski, et al.       Expires February 20, 2014               [Page 6]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   In our example :

   o  MPLS TE forwarding entry for T2 (ending on PE2) would be protected
      by TE-FRR (if enabled).

   o  IP forwarding entry for PE2 would be protected by LFA.

   In case of failure of C2-C3 :

   o  traffic from PE1 to PE2 (encapsulated in T2), would be protected
      by TE FRR.

   o  traffic from PE3 to PE2 (native IP), would be protected by LFA.

   In other words, if a node S has an IGP/LDP forwarding entry F1 with
   outgoing interface i1 and a MPLS TE midpoint forwarding entry F2 with
   outgoing interface i2, then an implementation MUST support using LFA
   FRR for F1 and TE FRR for F2 as long as i1 != i2.

   If i1 == i2, an implementation SHOULD allow for using LFA FRR backup
   for F1 and TE FRR backup for F2.

3.  Operational considerations

   In this section, we first discuss the benefit of considering a joint
   deployment of LFA and MPLS tunnels to achieve resiliency.  We then
   discuss one approach aiming at defining MPLS tunnels for the purpose
   of complementing LFA coverage.

3.1.  Relevance of joint LFA FRR and RSVP-TE FRR deployments

   This section describes the deployment scenarios where it can be
   beneficial to jointly use LFAs and RSVP-TE FRR.

   There are many networks where RSVP-TE is already deployed.  The
   deployment of RSVP-TE is typically for two main reasons :

   o  Traffic engineering : a provider wants to route some flows on some
      specific paths using constraints;

   o  Traffic protection using Fast-reroute ability

   LFA is a feature that may bring benefits on RSVP-TE enabled networks,
   with no/minimal operational cost (compared to RSVP-TE FRR global roll
   out).  These benefits include:






Litkowski, et al.       Expires February 20, 2014               [Page 7]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   o  Should increase protection on network where FRR is not available
      everywhere.  Although it may not provide full coverage, it will
      increase the protection significantly.

   o  May provide better protection in specific cases than RSVP-TE FRR

   For IP networks that do not have any traffic protection mechanism,
   LFA is a very good first step to provide traffic protection even if
   its coverage is not 100%. Providers may want to increase protection
   coverage if LFA benefit is not sufficient for some destinations, in
   some parts of the network.  The following sections discusses the use
   of basic RSVP-TE tunnels to extend protection coverage.

3.2.  Extending LFA coverage using RSVP-TE tunnels

   We already have seen in previous sections that RSVP-TE tunnels could
   be established by an operator to complement LFA coverage.  The method
   of tunnel placement depends on what type of protection (link or node)
   is required, as well as on the set of destinations or network parts
   which requires better protection than what LFA can provide.

3.2.1.  Creating multihop tunnel to extend topology

   To extend the coverage, the idea is to use a mechanism extending LFA
   by turning TE tunnels into LFA candidates.  This mechanism is of a
   local significance only.

   When explicitly establishing tunnels for that purpose, choices have
   to be made for the endpoints of such tunnels, in order to maximize
   coverage while preserving management simplicity.  Requirements are
   that:

   o  Endpoints must satisfy equations from [RFC5286], otherwise it will
      not be a valide LFA candidate: so when releasing traffic from
      tunnel, the traffic will go to the destination without flowing
      through the protected link or node.  Depending on which equations
      are satisfied, node or link protection will be provided by the
      tunnel hop.

   o  Tunnel must not flow though the link or node to be protected,
      explicit routing of tunnel is recommended to enforce this
      condition.

   The approach to choose tunnel endpoints might be different here when
   compared to [I-D.ietf-rtgwg-remote-lfa] as endpoint choice is a
   manual one.  Automatic behavior and scaling of
   [I-D.ietf-rtgwg-remote-lfa] requires:




Litkowski, et al.       Expires February 20, 2014               [Page 8]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   o  Non null intersection of Extended P-Space and Q-Space

   o  Computation of PQ node only for the remote end of the link

   Based on this, [I-D.ietf-rtgwg-remote-lfa] may:

   o  Not find a tunnel endpoint;

   o  Not provide the more efficient protection : -- i.e. provides only
      link protection, while there is node protection possible for a
      specific destination

   The proposed solution of manual explicitly routed tunnels is a good
   complement for [I-D.ietf-rtgwg-remote-lfa] and provides more
   flexibility:

   o  Always a possibility to find a tunnel endpoint for a specific
      destination.

   o  Possibility to provide a better protection type (link vs. node).

3.2.2.  Selecting multihop tunnels to extend topology

   From a manageability point of view, computing a best Q node for each
   destination could lead to have one different Q node for each
   destination.  This is not optimal in terms of number of tunnels,
   given that possibly one Q node may be able to serve multiple non
   covered destinations.

   Rather than computing the best Q node per non covered destination, we
   would prefer to find best compromise Q nodes (best for multiple
   destinations).  To find the best compromise between coverage increase
   and number of tunnels, we recommend to use a simulator performing the
   following computations per link:

   Step 1 :  Compute for each not covered destination (routed on the
      link) the list of endpoints that are satisfying equations from
      [RFC5286] (node or link protection equations depending of required
      level of protection) : nodes in Q-Space

   Step 2 :  Remove endpoints that are not eligible for repair (Edge
      nodes, low bandwidth meshed nodes, number of hops ...) : multiple
      attributes could be specified to exclude some nodes from Q-Space :
      The example of attributes include router type, metric to node,
      bandwidth, packet loss, RTD ...

   Step 3 :  Within the list of endpoints (one list per destination),
      order the endpoints by number of destination covered



Litkowski, et al.       Expires February 20, 2014               [Page 9]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   Step 4 :  Choose the endpoint that has the highest number of
      destination covered : some other criteria could be used to prefer
      an endpoint from another (same type of criteria that excluded some
      nodes from Q-Space)

   Step 5 :  Remove destinations covered by this endpoint from non
      covered list

   Step 6 :  If non covered list is not empty, restart from Step 1

   Multiple endpoint (and so tunnels) could be necessary to have 100%
   coverage.  But the idea is to find a tradeoff between number of
   tunnels configured (complexity) and number of destination covered,
   combining with traffic information would also provide a better view.

4.  Security Considerations

   TBD.

5.  Contributors

   Significant contribution was made by Pierre Francois which the
   authors would like to acknowledge.

6.  IANA Considerations

   This document has no actions for IANA.

7.  References

7.1.  Normative References

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119, March 1997.

   [RFC3906]  Shen, N. and H. Smit, "Calculating Interior Gateway
              Protocol (IGP) Routes Over Traffic Engineering Tunnels",
              RFC 3906, October 2004.

   [RFC4090]  Pan, P., Swallow, G., and A. Atlas, "Fast Reroute
              Extensions to RSVP-TE for LSP Tunnels", RFC 4090, May
              2005.

   [RFC5286]  Atlas, A. and A. Zinin, "Basic Specification for IP Fast
              Reroute: Loop-Free Alternates", RFC 5286, September 2008.

7.2.  Informative References




Litkowski, et al.       Expires February 20, 2014              [Page 10]

Internet-Draft           lfa-rsvpte-cooperation              August 2013


   [I-D.bryant-ipfrr-tunnels]
              Bryant, S., Filsfils, C., Previdi, S., and M. Shand, "IP
              Fast Reroute using tunnels", draft-bryant-ipfrr-tunnels-03
              (work in progress), November 2007.

   [I-D.ietf-rtgwg-remote-lfa]
              Bryant, S., Filsfils, C., Previdi, S., Shand, M., and S.
              Ning, "Remote LFA FRR", draft-ietf-rtgwg-remote-lfa-02
              (work in progress), May 2013.

Authors' Addresses

   Stephane Litkowski
   Orange

   Email: stephane.litkowski@orange.com


   Bruno Decraene
   Orange

   Email: bruno.decraene@orange.com


   Clarence FilsFils
   Cisco Systems

   Email: cfilsfil@cisco.com


   Kamran Raza
   Cisco Systems

   Email: skraza@cisco.com

















Litkowski, et al.       Expires February 20, 2014              [Page 11]