The objective of the label stack reduction is twofold:
It reduces the label stack so ingress PE routers with a lower Maximum SID Depth (MSD) can still work.
It provides the ability to spray packets over ECMP paths to an intermediate node SID when all these paths satisfy the constraints of the SR-TE LSP path. Even if the resulting label stack is not reduced, this aspect of the feature is still useful.
If the user enables the label-stack-reduction command for this LSP, a second phase is applied attempting to reduce the label stack that resulted from the fully explicit path with adjacency SIDs and adjacency sets SIDs computed in the first phase.
This is to attempt a replacement of adjacency and adjacency set SIDs corresponding to a segment of the explicit path with a node SID whenever the constraints of the path are met by all the ECMP paths to that node SID.
This is the procedure followed by the label stack reduction algorithm:
Phase 1 of the CSPF returns up to three fully explicit ECMP paths that are eligible for label stack reduction. These paths are equal cost from the point of view of IGP metric or TE metric as configured for that SR-TE LSP.
Each fully explicit path of the SR-TE LSP that is computed in Phase 1 of the CSPF is split into a number of segments that are delimited by the user-configured loose or strict hops in the path of the LSP. Label stack reduction is applied to each segment separately.
Label stack reduction in Phase 2 consists of traversing the CSPF tree for each ECMP path returned in Phase 1 and then attempting to find the farthest node SID in a path segment that can be used to summarize the entire path up to that node SID. This requires that all links of ECMP paths are able to reach the node SID from the current node on the CSPF tree to satisfy all the TE constraints of the SR-TE LSP paths. ECMP is based on the IGP metric, in this case, because this is what routers use in the data path when forwarding a packet to the node SID.
If the TE metric is enabled for the SR-TE LSP, then one of the constraints is that the TE metric must be the same value for all the IGP metric ECMP paths to the node SID.
CSPF in Phase 2 selects the first candidate ECMP path from Phase 1 which reduced label stack that satisfies the constraint carried in the max-sr-labels command.
The CSPF path computation in Phase 1 always avoids a loop over the same hop as is the case with the RSVP-TE LSP. In addition, the label stack reduction algorithm prevents a path from looping over the same hop because of the normal routing process. For example, it checks if the same node is involved in the ECMP paths of more than one segment of the LSP path and builds the label stack to avoid this situation.
During the MBB procedure of a timer or manual re-optimization of a SR-TE LSP path, the TE-DB performs additional steps as compared to the case of the initial path computation:
MPLS provides TE-DB with the current working path of the SR-TE LSP.
TE-DB updates the path’s metric based on the IGP or TE link metric (if TE metric enabled for the SR-TE LSP).
For each adjacency SID, it verifies that the related link and SID are still in its database and that the link fulfills the LSP constraints. If so, it picks up the current metric.
For each node SID, it verifies that the related prefix and SID are still available, and if so, checks that all the links on the shortest IGP path to the node owning the node SID fulfill the SR-TE LSP path constraints. This is re-using the same checks detailed in Step3 for the label compression algorithm.
CSPF computes a new path with or without label stack reduction as described in Steps 1, 2, and 3.
TE-DB returns both paths to MPLS. MPLS always programs the new path in the case of a manual re-optimization. MPLS compares the metric of the new path to the current path and if different, programs the new path in the case of a timer re-optimization.
TE-DB returns to MPLS the following information together with the reduced path ERO and label stack:
a list of SRLGs of each hop in the ERO represented by a node SID and that includes SRLGs used by links in all ECMP paths to reach that node SID from the previous hop.
the cost of each hop in the ERO represented by an adjacency SID or adjacency set SID. This corresponds to the IGP metric or TE metric (if TE is metric-enabled for the SR-TE LSP) of that link or set of links. In the case of an adjacency set, all TE metrics of the links must be the same, otherwise CSPF does not select the set.
the cost of each hop in the ERO represented by a node SID and this corresponds to the cumulated IGP metric or TE metric (if TE metric is enabled for the SR-TE LSP) to reach the node SID from the previous hop using the fully explicit path computed in Phase 1.
the total cost or computed metric of the SR-TE LSP path. This consists of the cumulated IGP metric or TE metric (if TE metric enabled for the SR-TE LSP) of all hops of the fully explicit path computed in Phase 1 of the CSPF.
If label stack reduction is disabled, the values of the max-sr-labels and the hop-limit commands are applied to the full explicit path in Phase 1.
The minimum of the two values is used as a constraint in the full explicit path computation.
If the resulting ECMP paths net hop-count in Phase 1 exceeds this minimum value no path is returned by TE-DB to MPLS
If label stack reduction is enabled, the values of the max-sr-labels and the hop-limit commands are both ignored in Phase 1 and only the value of the max-sr-labels is used as a constraint in Phase 2.
If the resulting net label stack size after reduction of all candidate paths in Phase 2 exceeds the value of parameter max-sr-labels then no path is returned by TE-DB to MPLS.
The label stack reduction does not support the use of an anycast SID, a prefix SID with N-flag clear, to replace a segment of the SR-TE LSP path. Only a node SID is used.