High Availability and Fast Convergence (NSR, NSF/GR, BFD, TI-LFA, PIC)
What this domain tests
High Availability and Fast Convergence tests control-plane redundancy and sub-second failure detection: NSR versus NSF/graceful restart, BFD timers, TI-LFA repair paths, and protection mechanisms such as LFA and BGP PIC.
NSF and NSR
- NSF/graceful restart preserves forwarding while the control plane restarts. Neighbors act as GR helpers and keep advertising the restarting router's routes.
- NSR synchronizes control-plane state to a standby RP and does not require neighbor support. BGP NSR also synchronizes TCP state.
- NSF can blackhole if the RIB/FIB is not preserved or the helper relationship fails; NSR avoids helper dependency but is platform- and protocol-specific.
BFD
- BFD async mode sends periodic control packets; detection time = interval × multiplier. Echo mode loops packets through the forwarding path.
- BFD can trigger OSPF, IS-IS, BGP, and static route failover faster than protocol hellos.
- BFD multihop is used for eBGP peers that are multiple hops away; micro-BFD monitors individual bundle members.
- Timers that are too aggressive cause false positives under CPU load.
TI-LFA
- TI-LFA uses segment routing to build a loop-free repair path for any topology.
- P-space is the set of nodes reachable from the source without the protected link; Q-space is the set of nodes that can reach the destination without the protected link; the PQ node is in both.
- The repair path is a label stack of adjacency and node SIDs, installed before failure.
PIC
- LFA and remote LFA provide local repair; BGP PIC edge pre-installs alternate eBGP paths; BGP PIC core protects IGP prefixes.
- MPLS TE FRR and MoFRR provide fast protection in their domains.
Worked example
R1 connects to R2 and R3; R2 and R3 connect to R4. The primary path is R1–R2–R4. When R1–R2 fails, R1 needs a repair path. P-space from R1 without R2 includes R1 and R3; Q-space to R4 without R2 includes R3 and R4; the PQ node is R3. R1 installs a repair label stack: adjacency SID to R3, then the node SID for R4. Traffic follows R1–R3–R4 until the IGP converges, without waiting for a new SPF.
Traps
- NSR is not available for every protocol or platform.
- NSF requires a helper and can fail if the restart is not graceful.
- BFD does not replace protocol hellos; it accelerates failure detection.
- TI-LFA requires SR and a valid PQ node; not every topology has one.
- LFA coverage is not 100 percent; remote LFA or TI-LFA extends it.