Network Working Group
Request for Comments: 3906
Category: Informational
N. Shen
Redback Networks
H. Smit
October 2004

Calculating Interior Gateway Protocol (IGP) Routes

Over Traffic Engineering Tunnels

Status of this Memo

This memo provides information for the Internet community. It does not specify an Internet standard of any kind. Distribution of this memo is unlimited.

Copyright Notice

Copyright © The Internet Society (2004).


This document describes how conventional hop-by-hop link-state routing protocols interact with new Traffic Engineering capabilities to create Interior Gateway Protocol (IGP) shortcuts. In particular, this document describes how Dijkstra's Shortest Path First (SPF) algorithm can be adapted so that link-state IGPs will calculate IP routes to forward traffic over tunnels that are set up by Traffic Engineering.

1. Introduction

Link-state protocols like integrated Intermediate System to Intermediate System (IS-IS) [1] and OSPF [2] use Dijkstra's SPF algorithm to compute a shortest path tree to all nodes in the network. Routing tables are derived from this shortest path tree. The routing tables contain tuples of destination and first-hop information. If a router does normal hop-by-hop routing, the first- hop will be a physical interface attached to the router. New traffic engineering algorithms calculate explicit routes to one or more nodes in the network. At the router that originates explicit routes, such routes can be viewed as logical interfaces which supply Label Switched Paths through the network. In the context of this document, we refer to these Label Switched Paths as Traffic Engineering tunnels (TE-tunnels). Such capabilities are specified in [3] and [4].

The existence of TE-tunnels in the network and how the traffic in the network is switched over those tunnels are orthogonal issues. A node may define static routes pointing to the TE-tunnels, it may match the recursive route next-hop with the TE-tunnel end-point address, or it may define local policy such as affinity based tunnel selection for switching certain traffic. This document describes a mechanism utilizing link-state IGPs to dynamically install IGP routes over those TE-tunnels.

The tunnels under consideration are tunnels created explicitly by the node performing the calculation, and with an end-point address known to this node. For use in algorithms such as the one described in this document, it does not matter whether the tunnel itself is strictly or loosely routed. A simple constraint can ensure that the mechanism be loop free. When a router chooses to inject a packet addressed to a destination D, the router may inject the packet into a tunnel where the end-point is closer (according to link-state IGP topology) to the destination D than is the injecting router. In other words, the tail-end of the tunnel has to be a downstream IGP node for the destination D. The algorithms that follow are one way that a router may obey this rule and dynamically make intelligent choices about when to use TE-tunnels for traffic. This algorithm may be used in conjunction with other mechanisms such as statically defined routes over TE-tunnels or traffic flow and QoS based TE- tunnel selection.

This IGP shortcut mechanism assumes the TE-tunnels have already been setup. The TE-tunnels in the network may be used for QoS, bandwidth, redundancy, or fastreroute reasons. When an IGP shortcut mechanism is applied on those tunnels, or other mechanisms are used in conjunction with an IGP shortcut, the physical traffic switching through those tunnels may not match the initial traffic engineering setup goal. Also the traffic pattern in the network may change with time. Some forwarding plane measurement and feedback into the adjustment of TE-tunnel attributes need to be there to ensure that the network is being traffic engineered efficiently [6].

2. Enhancement to the Shortest Path First Computation

During each step of the SPF computation, a router discovers the path to one node in the network. If that node is directly connected to the calculating router, the first-hop information is derived from the adjacency database. If a node is not directly connected to the calculating router, it inherits the first-hop information from the parent(s) of that node. Each node has one or more parents. Each node is the parent of zero or more down-stream nodes.

For traffic engineering purposes, each router maintains a list of all TE-tunnels that originate at this router. For each of those TE- tunnels, the router at the tail-end is known.

During SPF, when a router finds the path to a new node (in other words, this new node is moved from the TENTative list to the PATHS list), the router must determine the first-hop information. There are three possible ways to do this:

  • Examine the list of tail-end routers directly reachable via a TE-tunnel. If there is a TE-tunnel to this node, we use the TE-tunnel as the first-hop.
  • If there is no TE-tunnel, and the node is directly connected, we use the first-hop information from the adjacency database.
  • If the node is not directly connected, and is not directly reachable via a TE-tunnel, we copy the first-hop information from the parent node(s) to the new node.

The result of this algorithm is that traffic to nodes that are the tail-end of TE-tunnels, will flow over those TE-tunnels. Traffic to nodes that are downstream of the tail-end nodes will also flow over those TE-tunnels. If there are multiple TE-tunnels to different intermediate nodes on the path to destination node X, traffic will flow over the TE-tunnel whose tail-end node is closest to node X. In certain applications, there is a need to carry both the native adjacency and the TE-tunnel next-hop information for the TE-tunnel tail-end and its downstream nodes. The head-end node may conditionally switch the data traffic onto TE-tunnels based on user defined criteria or events; the head-end node may also split flow of traffic towards either types of the next-hops; the head-end node may install the routes with two different types of next-hops into two separate RIBs. Multicast protocols running over physical links may have to perform RPF checks using the native adjacency next-hops rather than the TE-tunnel next-hops.

3. Special Cases and Exceptions

The Shortest Path First algorithm will find equal-cost parallel paths to destinations. The enhancement described in this document does not change this. Traffic can be forwarded over one or more native IP paths, over one or more TE-tunnels, or over a combination of native IP paths and TE-tunnels.

A special situation occurs in the following topology:

      rtrA -- rtrB -- rtrC
               |       |
              rtrD -- rtrE

Assume all links have the same cost. Assume a TE-tunnel is set up from rtrA to rtrD. When the SPF calculation puts rtrC on the TENTative list, it will realize that rtrC is not directly connected, and thus it will use the first-hop information from the parent, which is rtrB. When the SPF calculation on rtrA moves rtrD from the TENTative list to the PATHS list, it realizes that rtrD is the tail- end of a TE-tunnel. Thus rtrA will install a route to rtrD via the TE-tunnel, and not via rtrB.

When rtrA puts rtrE on the TENTative list, it realizes that rtrE is not directly connected, and that rtrE is not the tail-end of a TE- tunnel. Therefore, rtrA will copy the first-hop information from the parents (rtrC and rtrD) to the first-hop information of rtrE. Traffic to rtrE will now load-balance over the native IP path via rtrA->rtrB->rtrC, and the TE-tunnel rtrA->rtrD.

In the case where both parallel native IP paths and paths over TE- tunnels are available, implementations can allow the network administrator to force traffic to flow over only TE-tunnels (or only over native IP paths) or both to be used for load sharing.

4. Metric Adjustment of IP Routes over TE-tunnels

   When an IGP route is installed in the routing table with a TE-tunnel
   as the next hop, an interesting question is what should be the cost
   or metric of this route?  The most obvious answer is to assign a
   metric that is the same as the IGP metric of the native IP path as if
   the TE-tunnels did not exist.  For example, rtrA can reach rtrC over
   a path with a cost of 20.  X is an IP prefix advertised by rtrC.  We
   install the route to X in rtrA's routing table with a cost of 20.
   When a TE-tunnel from rtrA to rtrC comes up, by default the route is
   still installed with metric of 20, only the next-hop information for
   X is changed.

While this scheme works well, in some networks it might be useful to change the cost of the path over a TE-tunnel, to make the route over the TE-tunnel less or more preferred than other routes.

For instance, when equal cost paths exist over a TE-tunnel and over a native IP path, by adjusting the cost of the path over the TE-tunnel, we can force traffic to prefer the path via the TE-tunnel, to prefer the native IP path, or to load-balance among them. Another example is when multiple TE-tunnels go to the same or different destinations. Adjusting TE-tunnel metrics can force the traffic to prefer some TE- tunnels over others regardless of underlining IGP cost to those destinations.

Setting a manual metric on a TE-tunnel does not impact the SPF algorithm itself. It only affects the comparison of the new route with existing routes in the routing table. Existing routes can be either IP routes to another router that advertises the same IP prefix, or it can be a path to the same router, but via a different outgoing interface or different TE-tunnel. All routes to IP prefixes advertised by the tail-end router will be affected by the TE-tunnel metric. Also, the metrics of paths to routers that are downstream of the tail-end router will be influenced by the manual TE-tunnel metric.

This mechanism is loop free since the TE-tunnels are source-routed and the tunnel egress is a downstream node to reach the computed destinations. The end result of TE-tunnel metric adjustment is more control over traffic loadsharing. If there is only one way to reach a particular IP prefix through a single TE-tunnel, then no matter what metric is assigned, the traffic has only one path to go.

The routing table described in this section can be viewed as the private RIB for the IGP. The metric is an important attribute to the routes in the routing table. A path or paths with lower metric will be selected over other paths for the same route in the routing table.

4.1. Absolute and Relative Metrics

It is possible to represent the TE-tunnel metric in two different ways: an absolute (or fixed) metric or a relative metric, which is merely an adjustment of the dynamic IGP metric as calculated by the SPF computation. When using an absolute metric on a TE-tunnel, the cost of the IP routes in the routing table does not depend on the topology of the network. Note that this fixed metric is not only used to compute the cost of IP routes advertised by the router that is the tail-end of the TE-tunnel, but also for all the routes that are downstream of this tail-end router. For example, if we have TE- tunnels to two core routers in a remote POP, and one of them is assigned with an absolute metric of 1, then all the traffic going to that POP will traverse this low-metric TE-tunnel.

By setting a relative metric, the cost of IP routes in the routing table is based on the IGP metric as calculated by the SPF computation. This relative metric can be a positive or a negative number. Not configuring a metric on a TE-tunnel is a special case of the relative metric scheme. No metric is the same as a relative metric of 0. The relative metric is bounded by minimum and maximum allowed metric values while the positive metric disables the TE- tunnel in the SPF calculation.

4.2. Examples of Metric Adjustment

Assume the following topology. X, Y, and Z are IP prefixes advertised by rtrC, rtrD, and rtrE respectively. T1 is a TE-tunnel from rtrA to rtrC. Each link in the network has an IGP metric of 10.

        ===== T1 =====>
      rtrA -- rtrB -- rtrC -- rtrD -- rtrE
           10      10  |   10  |   10  |
                       X       Y       Z

Without TE-tunnel T1, rtrA will install IP routes X, Y, and Z in the routing table with metrics 20, 30, and 40 respectively. When rtrA has brought up TE-tunnel T1 to rtrC, and if rtrA is configured with the relative metric of -5 on tunnel T1, then the routes X, Y, and Z will be installed in the routing table with metrics 15, 25, and 35. If an absolute metric of 5 is configured on tunnel T1, then rtrA will install routes X, Y, and Z all with metrics 5, 15, and 25 respectively.

5. Security Considerations

This document does not change the security aspects of IS-IS or OSPF. Security considerations specific to each protocol still apply. For more information see [5] and [2].

6. Acknowledgments

The authors would like to thank Joel Halpern and Christian Hopps for their comments on this document.

7. Informative References

   [1] ISO.  Information Technology - Telecommunications and Information
       Exchange between Systems - Intermediate System to Intermediate
       System Routing Exchange Protocol for Use in Conjunction with the
       Protocol for Providing the Connectionless-Mode Network Service.
       ISO, 1990.

[2] Moy, J., "OSPF Version 2", RFC 2328, April 1998.

[3] Awduche, D., Malcolm, J., Agogbua, J., O'Dell, M., and J.

McManus, "Requirements for Traffic Engineering Over MPLS", RFC 2702, September 1999.

[4] 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.

   [5] Li, T. and R. Atkinson, "Intermediate System to Intermediate
       System (IS-IS) Cryptographic Authentication", RFC 3567, July
   [6] Awduche, D., Chiu, A., Elwalid, A., Widjaja, I., and X. Xiao,
       "Overview and Principles of Internet Traffic Engineering", RFC
       3272, May 2002.

8. Authors' Addresses

Naiming Shen
Redback Networks, Inc.
300 Holger Way
San Jose, CA 95134


   Henk Smit


9. Full Copyright Statement

Copyright © The Internet Society (2004).

This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights.


Intellectual Property

The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the IETF's procedures with respect to rights in IETF Documents can be found in BCP 78 and BCP 79.

Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at

The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf-


Funding for the RFC Editor function is currently provided by the Internet Society.