RSVP-TE Mesh with Secondary Paths and Admin-Group Constraints
A 14-router MPLS core carries a full mesh of 56 RSVP-TE LSPs among eight PE routers. Each primary LSP requests 300 Mbps; core links are 10 Gbps with 30 percent reservable bandwidth. Voice LSPs carry an affinity that must exclude links colored 'no-voice'; the other LSPs have no affinity. Every LSP has a secondary path that is loosely routed through the core. Preemption is left at the default setup priority 7 and hold priority 0 on every LSP. After a fiber cut takes out two core links, several primary LSPs resignal, but a subset of the secondaries never leave the 'CSPF failed' state, and one voice LSP ends up on a link it was supposed to avoid. Some LSPs also show 'Deselected' with a large ResvTear count. No LSP has auto-bandwidth enabled.
1. Explain why some secondary LSPs cannot be resignaled after the fiber cut, given that the core still has free capacity on other links. What preemption and bandwidth changes would you make, and what is the risk of each?
Read the worked answer
CSPF failure after a failure usually means the path computation cannot find a route that satisfies both bandwidth and affinity, not that the core is full. Each secondary requests 300 Mbps. After the cut, the remaining links may have enough aggregate bandwidth but not a single path with 300 Mbps of unreserved bandwidth, or the loose path may traverse a link excluded by affinity. Preemption makes it worse: setup priority 7 is the lowest, so a new LSP cannot preempt any existing LSP; hold priority 0 is the highest, so established LSPs are hard to preempt. A secondary that needs to resignal after a failure is treated as a new setup and cannot displace the primaries that are now occupying the surviving links.
Set consistent priorities by class: voice setup 3 / hold 2, business setup 5 / hold 4, best-effort setup 7 / hold 7. This lets voice preempt data if necessary, while data cannot preempt voice. Also check show rsvp interface for available bandwidth, show ted database extensive for link reservations, and show mpls lsp extensive for the CSPF error. Risk: aggressive preemption causes churn and can move a live LSP; use it only for the classes that justify it. If the topology has a single bottleneck, add capacity or reduce the requested bandwidth; preemption cannot create bandwidth.
2. A voice LSP ends up on a link it should avoid. Walk through how CSPF evaluates admin groups/affinity for primary and secondary paths, and what you would check to find the missing constraint.
Read the worked answer
CSPF compares the LSP's affinity with each link's admin groups. include-any requires at least one listed group; include-all requires all; exclude rejects any link carrying the listed group. The voice LSP should have exclude no-voice on both primary and secondary. A common failure is to apply the affinity only to the primary path, so after a failure the secondary CSPF ignores the constraint and selects the no-voice link. Another is that the link's admin group was never configured, so the exclusion matches nothing. Check show configuration protocols mpls label-switched-path <lsp>, show rsvp interface detail, and show ted database extensive | match admin. Verify the secondary path with show mpls lsp extensive and look at the actual path. If the LSP is loosely routed, CSPF can choose any link that satisfies the affinity; make the secondary explicit if the constraint is critical.
3. The LSPs use make-before-break. Explain why a resignal can temporarily require double bandwidth and how that interacts with the failed secondaries. What tuning would you apply?
Read the worked answer
Make-before-break signals a new path with the same tunnel ID but a different LSP ID, and keeps the old path until the new one is established. During that window, the new path reserves bandwidth while the old path still holds its reservation, so the headend needs twice the bandwidth. After a fiber cut, surviving links are already carrying the failed LSPs' resigned paths and the secondaries; an MBB resignal of a primary can therefore fail CSPF even though the steady-state demand fits. The 'Deselected' state and large ResvTear count are symptoms of repeated resignal attempts and teardown of paths that lose the comparison.
Tune resignal-timer and optimize-timer to avoid immediate reoptimization during convergence, and do not enable aggressive auto-bandwidth at the same time. Use standby on secondaries only where the class requires it; standby reserves bandwidth and increases the double-counting problem. Verify with show mpls lsp extensive for the path history and show rsvp session detail for the reservations. If the core cannot support MBB during a single failure, reduce LSP bandwidth or add diversity.