A large service provider network has a single-level IS-IS network with 500 routers. This network has short- haul and long-haul links. Periodically, long-haul links bounce for a short period of time, c

Modifying the LSP lifetime controls how long LSPs remain in the database before requiring a refresh, which has no direct effect on CPU load during rapid link flapping events.

Modifying hello timers on short-haul link routers changes neighbor adjacency detection speed on those links and does not address the CPU overload caused by flapping long-haul links elsewhere.

LSP pacing (transmission throttling) limits how rapidly a router floods LSP packets to its neighbors, so when a long-haul link flaps 10-20 times in a few minutes the older routers are not overwhelmed by continuous flooding cycles. Because pacing controls only the transmission rate and not LSP generation, the very first LSP for any new topology change is still generated and sent immediately, preserving fast convergence for all interface failures. Applying pacing selectively to routers with long-haul links targets the source of the excessive flapping without affecting short-haul link behavior.

LSP generation throttling delays how quickly a new LSP is created after a topology change, which directly slows down initial convergence for all interface failures and violates the requirement of no impact on fast convergence.