TI-LFA (Topology Independent Loop-Free Alternate)¶

Topology Independent Loop-Free Alternate — TI-LFA, sometimes written TILFA or "TI LFA" — is the fast-reroute (FRR) mechanism of Segment Routing. Every router pre-computes, for each destination, a backup path that follows the route the network will use after the failure, and encodes it as a short segment list. When a link or neighbor fails, the router switches traffic onto that list in the data plane, without waiting for the IGP. It is standardized in RFC 9855 (October 2025, Proposed Standard) and works identically for SR-MPLS and SRv6.

TI-LFA builds on the older IP FRR ideas — LFA, Remote LFA and Directed LFA — and, in the RFC's words, extends them to "guaranteed coverage in any two-connected network". Segment Routing is what makes that possible: the repair router can steer a packet along any path with a couple of SIDs, so it never depends on a lucky neighbor.

The Problem: The Convergence Window¶

When a link fails, the IGP must detect the failure, flood the change, run SPF and update the FIB on every affected router. Until that finishes — hundreds of milliseconds to seconds in a large network — packets toward the failed link are dropped or loop.

IP FRR closes that gap locally: the router next to the failure, the Point of Local Repair (PLR), has a backup next hop pre-installed in its FIB and switches to it as soon as it detects the failure (loss of light, BFD). The question is where that backup path comes from.

Key Terms¶

Worked Example: Finding P, Q and the Repair List¶

Six routers, IS-IS metrics on the links. S is the PLR, protecting its link to N, for traffic to D.

graph LR
    S((S)) ---|10 · protected| N((N))
    N ---|10| D((D))
    S ---|10| A((A))
    A ---|10| B((B))
    B ---|40| C((C))
    C ---|10| D

Before the failure, S reaches D via N (cost 20). A also reaches D via S and N (cost 30), so A is not an LFA: by the RFC 5286 inequality, A is loop-free only if dist(A,D) < dist(A,S) + dist(S,D), i.e. 30 < 10 + 20, which is false. Classic LFA has nothing to offer here.

Step 1 — post-convergence path. Remove S–N and recompute: S → A → B → C → D, cost 70.

Step 2 — P-space of S (avoiding S–N).

Extended P-space adds what A can reach without S–N: B, but not C (A's shortest path to C, cost 40, goes back through S–N). Extended P-space = {A, B}.

Step 3 — Q-space of D (avoiding S–N).

Step 4 — intersect along the post-convergence path. P nodes on the path: A, B. Q nodes on the path: C. There is no PQ node, but the last P node (B) and the first Q node (C) are adjacent. RFC 9855 §5.3 gives the repair list:

RL = < Node-SID(B), Adj-SID(B→C) >

The adjacency SID is essential: if S only sent the packet to B, B would forward it on its own shortest path to D — back through A and S — and the packet would loop. The Adj-SID forces B to use the expensive B–C link, after which C's normal routing delivers it.

In SRv6 the same list becomes SRv6 SIDs: B's End (uN) SID followed by B's End.X (uA) SID toward C. With uSID both fit in one 128-bit container, so the PLR adds a single outer IPv6 header and no SRH:

Original packet at S:   IPv6(DA = D's service SID) | payload
Repaired at S:          IPv6(DA = fcbb:bb00:B:BC::) | IPv6(DA = D's service SID) | payload
                                    └uN(B) └uA(B→C)
At B:  uN shifts, uA forwards over B–C and, with PSP/USD, removes the outer header
At C:  plain IPv6 forwarding of the original packet to D

Repair-list shapes (RFC 9855 Section 5)¶

How often each case occurs is measured in RFC 9855 Appendix B on nine real networks: for link and local-SRLG protection, one SID or fewer covers more than 99% of destinations; for node protection, two SIDs or fewer cover 99%. On SRv6, adjacency SIDs are globally routable, so the list is often even shorter than on SR-MPLS (§7.2).

Link vs Node vs SRLG Protection¶

The algorithm is the same; what changes is the resource X removed from the topology before computing the post-convergence path, P-space and Q-space.

Two things are easy to miss:

• A link-protecting backup can be useless for a node failure. If S has two parallel links to N, the link-protecting backup for one of them is simply the other — which dies too if N reboots. Node protection forces the list around N. If no node-protecting path exists, implementations fall back to link protection.
• Node protection only protects transit through N. Traffic whose destination is N (or a service SID owned by N) cannot be rerouted around N; RFC 9855 §6.1 keeps an active node segment on the packet and adds the repair list in front of it. Protecting traffic to a failed egress PE is a separate problem (see draft-ietf-rtgwg-srv6-egress-protection and BGP PIC).

Vendors pick between the modes with tiebreakers. A node-protecting path is preferred only when you configure it. Check SRLG support per data plane: Junos ignores SRLG in SRv6 TI-LFA and does not support SRLG protection in IPv6 networks.

How TI-LFA Works with SRv6¶

RFC 9855 §7.2 states that the computation is identical for SRv6; only the encoding differs:

• At a transit PLR, the repair list is added with a headend behavior from RFC 8986 §5 — typically H.Encaps.Red, an outer IPv6 header whose DA is the first repair SID.
• At a PLR that is itself a segment endpoint for the packet, it first executes its endpoint behavior and then protects the packet like a transit packet.
• Repair SIDs advertised with the USD (and PSP) flavor let the last repair node remove the outer header and forward the original packet. That is why Cisco shows the P node's SID as uN (PSP/USD).

Implementations differ in how the repair is applied, so don't assume one behavior:

Configuration¶

segment-routing
 srv6
  locators
   locator MAIN
    micro-segment behavior unode psp-usd
    prefix fcbb:bb00:1::/48
!
router isis CORE
 address-family ipv6 unicast
  metric-style wide
  microloop avoidance segment-routing
  segment-routing srv6
   locator MAIN
 !
 interface HundredGigE0/0/0/1
  point-to-point
  address-family ipv6 unicast
   fast-reroute per-prefix
   fast-reroute per-prefix ti-lfa
   !! optional: prefer node-protecting, then SRLG-disjoint backups
   fast-reroute per-prefix tiebreaker node-protecting index 100
   fast-reroute per-prefix tiebreaker srlg-disjoint index 200

!! SR-MPLS TI-LFA
router isis
 fast-reroute per-prefix level-2 all
 fast-reroute ti-lfa level-2
 fast-reroute tie-break level-2 node-protecting 100
!
interface GigabitEthernet0/0/1
 isis network point-to-point
!
!! SRv6 TI-LFA (IOS XE 17.15.1a and later)
segment-routing srv6
 locators
  locator A
   prefix CAFE:0:601::/48
   format usid-f3216
!
router isis 1
 address-family ipv6
  multi-topology
  segment-routing srv6
   locator A
  fast-reroute per-prefix level-2 all
  fast-reroute ti-lfa level-2
 exit-address-family

# TI-LFA for SR-MPLS and SRv6 in IS-IS
set protocols isis backup-spf-options use-post-convergence-lfa
set protocols isis backup-spf-options use-post-convergence-lfa maximum-backup-paths 2
set protocols isis backup-spf-options use-source-packet-routing
set protocols isis interface et-0/0/1.0 level 2 post-convergence-lfa
# optional, per interface
set protocols isis interface et-0/0/1.0 level 2 post-convergence-lfa node-protection
set protocols isis interface et-0/0/1.0 level 2 post-convergence-lfa srlg-protection   # SR-MPLS/IPv4 only

# SRv6 microloop avoidance
set protocols isis spf-options microloop-avoidance post-convergence-path delay 5000

Verification¶

show isis ipv6 fast-reroute fcbb:bb00:2::/48 detail
!! Backup path: TI-LFA (link), via fe80::..., HundredGigE0/0/0/1 Node3, Metric: 40
!!   P node: Node4.00 [...], SRv6 SID: fcbb:bb00:4:: uN (PSP/USD)

show route ipv6 fcbb:bb00:2::/48 detail
!!   ... Backup (TI-LFA)   Repair Node(s): ...
!!   SRv6 Headend: H.Encaps.Red, SID-list {fcbb:bb00:4::}

show cef ipv6 fcbb:bb00:2::/48 detail
show cef ipv6 fast-reroute-db

show isis fast-reroute ti-lfa tunnel          ! SR-MPLS repair tunnels
show ip cef 10.0.0.2 internal
show isis ipv6 fast-reroute summary            ! SRv6: protected prefixes and coverage
show isis ipv6 rib fcbb:bb00:2::/48            ! "repair path: via ... (DS,NP)"
show ipv6 cef fcbb:bb00:2::/48 internal        ! "Segment List ... mode:insert"
show isis ipv6 fast-reroute ti-lfa fwd-ids

show isis backup coverage            # % of nodes and prefixes protected per address family
show isis backup spf results         # backup computation per destination
show isis backup spf results no-coverage
show route 2001:db8::2 extensive     # primary and backup next hops

TI-LFA and Micro-Loops¶

TI-LFA protects the moment of failure at the PLR. It does nothing for the micro-loops that can form elsewhere while routers update their FIBs at different times, or when the link comes back. RFC 9855 §1 calls the two mechanisms complementary and warns of one interaction: if the PLR converges early and releases its TI-LFA path before its neighbors have converged, the packets that were safely on the repair path can start looping — especially for repairs built from §5.2–5.4 (PQ or P/Q nodes rather than a direct neighbor).

The usual fix is to keep the TI-LFA path in place for a while after local convergence (RFC 8333-style local delay) or to steer traffic over a loop-free SR path until the network settles:

router isis CORE
 address-family ipv6 unicast
  microloop avoidance segment-routing
  microloop avoidance rib-update-delay 5000

On IOS-XR the delay range is documented as 1–60000 ms with a default of 5000. See Micro-Loop Avoidance for the full picture.

TI-LFA vs Other FRR Mechanisms¶

Further Reading¶

• RFC 9855 - Topology Independent Fast Reroute Using Segment Routing
• Micro-Loop Avoidance - Preventing loops during the convergence window after TI-LFA
• uSID / SRv6 Compression - Efficient backup segment encoding
• SRH Mechanics & Packet Walk - How backup segments are processed
• Flex-Algorithm - TI-LFA within a Flex-Algo topology
• Mission-Critical Networks - TI-LFA with node/SRLG/linecard-disjoint tiebreakers in a validated defense design
• RFC 9860 - MoFRR based on TI-LFA (multicast protection)
• Cisco IOS-XR - SRv6 configuration on IOS-XR
• Cisco IOS-XE - SRv6 configuration on IOS-XE
• Juniper - SRv6 configuration on Junos

References¶

1. RFC 9855 - Topology Independent Fast Reroute Using Segment Routing - Standards-track definition of TI-LFA: P/Q-space, repair lists, SR-MPLS and SRv6 data planes, coverage analysis
2. RFC 7490 - Remote Loop-Free Alternate (LFA) Fast Reroute (FRR) - Remote LFA and the PQ node definition, with coverage results
3. RFC 6571 - LFA Applicability in Service Provider Networks - Measured LFA coverage across 11 service-provider topologies
4. RFC 5286 - Basic Specification for IP Fast Reroute: Loop-Free Alternates - The original LFA inequalities
5. RFC 8333 - Micro-loop Prevention by Introducing a Local Convergence Delay - Delaying local FIB updates so FRR paths stay in place
6. Cisco 8000: Configure TI-LFA for SRv6 - IOS-XR SRv6 TI-LFA configuration and verification output
7. Cisco 8000: Usage Guidelines for SRv6 TI-LFA - Tiebreaker priorities, node and SRLG protection rules
8. Cisco IOS-XE: IS-IS SRv6 TI-LFA - SRv6 TI-LFA on IOS XE 17.15.1a and later
9. Cisco IOS-XE: IS-IS TI-LFA FRR - SR-MPLS TI-LFA configuration, tiebreakers and verification
10. Cisco: IS-IS SRv6 Microloop Avoidance Behavior - TI-LFA and microloop avoidance sequence on NCS platforms (H.Insert.Red repair)
11. Juniper: Configuring TI-LFA Using SPRING for IS-IS - Junos TI-LFA configuration example
12. Juniper: Understanding TI-LFA for IS-IS - Junos backup options, limits and IPv6 caveats
13. Juniper: SRv6 Network Programming in IS-IS - SRv6 TI-LFA on Junos, transit SRH insertion and the SRLG limitation
14. RFC 9860 - Multicast-Only Fast Reroute (MoFRR) Based on TI-LFA - Uses TI-LFA repair paths and PIM RPF Vectors to build MoFRR standby trees