Convergence
Convergence
This domain tests how fast OSPF reacts to change and which timers and features control that reaction. Questions usually ask what to tune and what the trade-off is.
Core rules
- Failure detection is driven by the dead interval; sub-second hellos (for example hello 1, dead 3) reduce detection time but raise CPU and bandwidth use.
- BFD gives faster, protocol-independent detection and is enabled per OSPF interface. It is the preferred tool when the dead interval is kept conservative.
- LSA generation and SPF runs are throttled: an initial delay, a hold time and a maximum wait. This prevents CPU spikes during flapping but adds delay to the first change.
- Graceful restart lets a control plane restart while forwarding is preserved, using the helper router to keep the adjacency and refresh LSAs.
- Stub and NSSA design reduces LSA count and therefore SPF scope, which indirectly speeds convergence.
Worked example
A link flaps ten times in a minute. Without throttling, each flap triggers a full SPF and floods. With SPF timers set to a short initial delay, a growing hold time up to a maximum, the later flaps are batched into fewer SPF runs while the first failure is still handled quickly.
Traps that cost marks
- Setting hello and dead very low across a large area and causing CPU problems.
- Enabling BFD only on one side of a link.
- Expecting graceful restart to help after a hardware failure; it covers control-plane restart only.
- Forgetting that timer values must match between neighbours.