Loop-free alternates

SR Linux supports the node-protect extensions to the remote LFA algorithm, as described in RFC 8102. When node protection is enabled, the router prefers a node-protect over a link-protect repair tunnel in the RLFA SPF computations. This feature protects against the failure of a downstream node. The node-protect extensions are additions to the original link-protect LFA SPF algorithm.

Remote LFA follows a similar algorithm as TI-LFA but does not limit the scope of the calculation of the extended P-Space and of the Q-Space to the post-convergence paths.

Remote LFA adds an extra forward SPF on behalf of the PQ node to ensure that, for each destination, the selected PQ-node does not use a path via the protected node.

The following figure shows an example of remote LFA for node protection.

Using the topology in the preceding figure, the node-protect remote LFA algorithm computation is performed in the following sequence.

1. Compute extended P-Space of R1 with respect to protected node R2.

   This is the set of nodes Yi which are reachable from R1 neighbors, other than protected node R2, without any path transiting the protected node R2.

   R1 computes an LFA SPF rooted at each of its neighbors, for example, R3, using the following equation:

   Distance_opt(R3, Yi) < Distance_opt(R3, R2) + Distance_opt(R2, Yi)

   Where Distance_opt(A,B) is the shortest distance between A and B.

   Nodes R3, R5, and R6 satisfy this inequality.

2. Compute Q-space of R1 with respect to protected link R1-R2.

   This is the set of nodes Zi from which node R2 can be reached without any path transiting the protected link R1-R2, using the following equation.

   Distance_opt(Zi, R2) < Distance_opt(Zi, R1) + Distance_opt(R1, R2)

   The reverse SPF for the Q-space calculation is the same as in the remote LFA link-protect algorithm and uses the protected node R2 as the proxy for all destination prefixes.

   This step yields nodes R3, R4, R5, and R6.

3. For each PQ node found, run a forward SPF to each destination D.

   This step is required to select only the subset of PQ-nodes that do not traverse protected node R2.

   Distance_opt(PQi, D) < Distance_opt(PQi, R2) + Distance_opt(R2, D)

   Of the candidates PQ nodes R3, R5, and R6, only PQ-nodes R5 and R6 satisfy this inequality. PQ-node R6 is closer in terms of IGP cost to computing router R1 and is therefore selected.

   This step of the algorithm is applied to the subset of candidate PQ-nodes out of steps 1 and 2 and to which the max-pq-cost parameter was already applied. This subset is further reduced in this step by retaining the candidate PQ-nodes that provide the highest coverage among all protected nodes in the topology, and the number of which does not exceed the value of the max-pq-nodes parameter.

4. Select a PQ-Node.

   If multiple PQ nodes satisfy the criteria in all the above steps, then R1 further selects the PQ node as follows.

   1. R1 selects the lowest IGP cost from R1.

   2. If more than one PQ-nodes remains, R1 selects the PQ-node reachable via the neighbor with the lowest system ID.

   3. If more than one PQ-node remains, R1 selects the PQ node with the lowest system ID.

5. Program the remote LFA backup path.
   For each destination prefix D, R1 programs the remote LFA backup path as follows.

   • For prefixes of R4 or downstream of R4, R1 programs a node-protect remote LFA repair tunnel to the PQ-node R6 by pushing the SID of node R6 on top of the SID for destination D and programming a next hop of R3.

   • For prefixes owned by node R2, R1 runs the link-protect remote LFA algorithm and programs a simple link-protect repair tunnel that consists of a backup next hop of R3 and pushes no additional label on top of the SID for the destination prefix.

   • Prefixes owned by nodes R3, R6, and R5 are not impacted by the failure of R2 because their primary next hop is R3.

For each destination node or prefix D, the TI-LFA link protection algorithm searches on each of the post-convergence paths for the closest Q-node to the computing node and then selects the closest P-node to this Q-node, up to a maximum configurable number of 3 labels.

The following figure shows a topology where router R3 computes a TI-LFA next hop for protecting link R3-R4.

In this topology, router R3 computes the link-protected TI-LFA backup path in the following order.

1. Compute the post-convergence SPF on the topology without the protected link.

   R3 finds a single post-convergence path to destination D via R1.

2. Compute the extended P-Space of R3 with respect to protected link R3-R4 on the post-convergence paths.

   This is the set of nodes Yi in the post-convergence paths that are reachable from R3 neighbors without any path transiting the protected link R3-R4.

   R3 computes an LFA SPF rooted at each of its neighbors within the post-convergence paths, that is, R1, using the following equation:

   Distance_opt(R1, Yi) < Distance_opt(R1, R3) + Distance_opt(R3, Yi)

   Where Distance_opt(A,B) is the shortest distance between A and B. The extended P-space calculation yields only node R1.

3. Compute the Q-space of R3 with respect to protected link R3-R4 in the post-convergence paths.

   This is the set of nodes Zi in the post-convergence paths from which the neighbor node R4 of the protected link, acting as a proxy for all destinations D, can be reached without any path transiting the protected link R3-R4.

   Distance_opt(Zi, R4) < Distance_opt(Zi, R3) + Distance_opt(R3, R4)

   The Q-space calculation yields nodes R2 and R4.

   This is the same computation of the Q-space performed by the remote LFA algorithm, except that the TI-LFA Q-space computation is performed only on the post-convergence paths.

4. For each post-convergence path, search for the closest Q-node and select the closest P-node to this Q-node, up to the configurable maximum number of labels.

   The preceding topology diagram shows a single post-convergence path and a single P-node (R1). It also shows that R2 is the closest of the two found Q-nodes to the P-Node.

   R3 installs the repair tunnel to the P-Q set and includes the node SID of R1 and the adjacency SID of the adjacency over link R1-R2 in the label stack. Because the P-node R1 is a neighbor of the computing node R3, the node SID of R1 is not needed and the label stack of the repair tunnel is compressed to the adjacency SID over link R1-R2 as shown.

   When a P-Q set is found on multiple ECMP post-convergence paths, the following selection rules are applied, in ascending order, to select a set from a single path:

   1. the lowest number of labels

   2. the next hop to the neighbor router with the lowest System ID

   3. the next hop corresponding to the Q node with the lowest System ID

   If multiple links with adjacency SIDs exist between the selected P-node and the selected Q-node, the following rules are used for link selection:

   1. the adjacency SID with the lowest metric

   2. the adjacency SID with the lowest SID value if the lowest metric is the same

Depending on the LFA options that are configured, the LFA SPF in the IGP instance runs the following algorithms in order.

1. Compute a regular LFA for each node and prefix.

   In this step, a computed backup next hop satisfies any applied LFA policy. This backup next hop protects that specific prefix or node in the context of SR FRR, and TE-Policy SR-MPLS FRR.

2. If TI-LFA is enabled, run the TI-LFA algorithm for all prefixes and nodes.

   If the loopfree-alternate option is enabled, a prefix or node for which a TI-LFA backup next hop is found overrides the result from step 1 in the context of SR FRR and in TE-Policy SR-MPLS FRR.

   With SR FRR, the TI-LFA next hop protects the node-SID of that prefix and any adjacency SID terminating on the same node-SID.

3. If RLFA is enabled, run remote LFA only for the next hop of prefixes and nodes that remain unprotected after steps 1 and 2.

   A prefix or node for which a remote LFA backup next hop is found uses it in the context of SR FRR and in TE-Policy SR-MPLS FRR.

SR Linux supports the node-protect extensions to the TI-LFA algorithm as described in draft-ietf-rtgwg-segment-routing-ti-lfa-01. When node protection is enabled, the router prefers a node-protect over a link-protect repair tunnel in the TI-LFA SPF computations. This feature protects against the failure of a downstream node. The node-protect extensions are additions to the original link-protect LFA SPF algorithm.

The following figure shows a simple topology to illustrate the node-protect operation for TI-LFA. It uses the same topology as shown in Remote LFA node-protect operation, but the metric for link R3-R6 is modified to 1000.

The main change as a result of the node-protect extension is that the algorithm protects a node instead of a link.

Using the topology in the preceding figure, the node protection computation is performed in the following sequence.

1. Compute the post-convergence SPF on the topology without the protected node.

   In the preceding figure, R1 computes TI-LFA on the topology without the protected node R2 and finds a single post-convergence path to destination D via R3 and R6.

   Prefixes owned by all other nodes in the topology have a post-convergence path via R3 and R6 except for prefixes owned by node R2. The latter uses the link R3-R2 and they can only benefit from link protection.

2. Compute the extended P-Space of R1 with respect to protected node R2 on the post-convergence paths.

   This is the set of nodes Yi in the post-convergence paths that are reachable from R1 neighbors, other than protected node R2, without any path transiting the protected node R2.

   R1 computes an LFA SPF rooted at each of its neighbors within the post-convergence paths. For example, R1 uses using the following calculation to compute the LFA SPF for R3:

   Distance_opt(R3, Yi) < Distance_opt(R3, R2) + Distance_opt(R2, Yi)

   Where Distance_opt(A,B) is the shortest distance between A and B.

   The extended P-space calculation yields node R3 only.

3. Compute the Q-space of R1 with respect to protected link R1-R2 on the post-convergence paths.

   This is the set of nodes Zi in the post-convergence paths from which node R2 can be reached without any path transiting the protected link R1-R2, using the following equation:

   Distance_opt(Zi, R2) < Distance_opt(Zi, R1) + Distance_opt(R1, R2)

   The reverse SPF for the Q-space calculation is the same as in the link-protect algorithm and uses the protected node R2 as the proxy for all destination prefixes. To compute the Q space with respect to the protected node R2 instead of link R1-R2, a reverse SPF would have to be performed for each destination D which is very costly and not scalable. However, this means the path from the Q-node to the destination D or the path from the P-node to the Q-node is not guaranteed to avoid the protected node R2. The intersection of the Q-space with post-convergence path is modified in the next step to mitigate this risk.

   This step yields nodes R3, R4, R5, and R6.

4. For each post-convergence path, search for the closest Q-node to destination D and select the closest P-node to this Q-node, up to the configurable maximum number of labels.

   This step yields the following P-Q sets, depending on the value of the ti-lfa max-sr-policy-lfa-labels parameter:

   • max-sr-policy-lfa-labels=0

     R3 is the closest Q-node to the destination D and R3 is the only P-node. This case results in link protection via PQ-node R3.

   • max-sr-policy-lfa-labels=1

     R6 is the closest Q-node to the destination D and R3 is the only P-node. The repair tunnel for this case uses the SID of the adjacency over link R3-R6 and is shown in the topology diagram.

   • max-sr-policy-lfa-labels=2 (default)

     R5 is the closest Q-node to the destination D and R3 is the only P-node. The repair tunnel for this case uses the SIDs of the adjacencies over links R3-R6 and R6-R5.

   • max-sr-policy-lfa-labels=3

     R4 is the closest Q-node to the destination D and R3 is the only P-node. The repair tunnel for this case uses the SIDs of the adjacencies over links R3-R6, R6-R5, and R5-R4.

   This step of the algorithm is modified from link protection, which prefers Q-nodes that are the closest to the computing router R1. This is to minimize the probability that the path from the Q-node to the destination D, or the path from the P-node to the Q-node, goes via the protected node R2 as described in step 2. However, there is still a probability that the found P-Q set achieves link protection only.

5. Select the P-Q Set.

   If a candidate P-Q set is found on each of the multiple ECMP post-convergence paths in step 4, the following selection rules are applied in ascending order to select a single set:

   1. the lowest number of labels

   2. the lowest next-hop router ID

   3. the lowest subinterface index if the same as the next-hop router ID

   If multiple parallel links with adjacency SID exist between the P- and Q-nodes of the selected P-Q set, the following rules are used to select one of them:

   1. the adjacency SID with lowest metric

   2. the adjacency SID with the lowest SID value, if the same as the lowest metric

6. Program the TI-LFA repair tunnel.

   For each destination prefix D, R1 programs the TI-LFA repair tunnel (max-sr-policy-lfa-labels=1):

   • For prefixes other than those owned by node R2, R1 programs a node-protect repair tunnel to the P-Q pair R3-R6 by pushing the SID of adjacency R3-R6 on top of the SID for destination D and programming a next hop of R3.

   • For prefixes owned by node R2, R1 runs the link-protect TI-LFA algorithm and programs a simple link-protect repair tunnel, which consists of a backup next hop of R3 and pushes no additional label on top of the SID for the destination prefix.