Fault-Tolerant IP and MPLS Networks

Comparison of the IS-IS Restart Mechanisms 131

5.

To have the neighbor set SRMflags for all LSPs (ensuring database synchronization), R2

transmits the CSNP containing the special LSP. The special CSNP is transmitted as a

single fragment.

6. Upon receipt of the CSNP (covering the entire range of LSP IDs) and containing the

special LSP, R1 sets SRMflags for all LSPs. In addition, R1 transmits a PSNP requesting

the special LSP from R2.

7. After sending the CSNP containing the special LSP and having received its confirmation

through the PSNP requesting the special LSP, R2 sends a set of normal CSNPs so that R1

does not unnecessarily transmit those LSPs that it already possesses.

8. Upon receiving a normal set of CSNPs, R1 clears the SRMflags for those LSPs that R2

already claims to have. For the remaining LSPs with SRMflags set, R1 transmits the

corresponding LSPs.

9. Following normal procedures, R2 synchronizes its database. Because R2 does not have

any reliable method to determine when its database is complete, it uses heuristics such as

a timer or waits until the LSPs’ rate of arrival falls below a certain threshold. Upon

completion of the database synchronization, R2 runs the SPF algorithm and updates the

FIB tables. To avoid causing a routing flap in other routers, R2 regenerates its own LSPs,

which are the same as those generated before this restart. Therefore, reception of R2’s

LSPs should not trigger SPF calculations on R1 and other routers in the domain.

10. Normal operation resumes.

Comparison of the IS-IS Restart Mechanisms

The advantages of the IETF IS-IS restart approach include scalability (does not require

IS-IS–related state preservation across the restart) and standards-based implementation.

Disadvantages of this approach are that it requires IS-IS protocol extensions and support from

its neighbors (helper nodes). To elaborate on this further, consider a router R1 in a provider’s

network that is peering with an R2 that might be managed by another provider or a customer.

The IETF IS-IS restart mechanism requires that both R1 and R2 support this capability.

However, the customer or the network provider might not always be willing to upgrade, or R2

might not be upgradeable.

An advantage of the Cisco IS-IS restart approach is that it does not require support from

neighbors. Therefore, unlike the IETF IS-IS restart, a Cisco IS-IS restart- and NSF-capable

router handles the case, whereas its neighbors are not IS-IS restart-capable. However, Cisco

IS-IS restart might be less scalable than the IETF IS-IS restart, because some IS-IS-related state

information must be preserved across the restart. Another drawback of this approach is the lack

of a reliable mechanism to determine when database synchronization is complete. Note that the

Cisco IOS architecture supports both IETF and Cisco-specific IS-IS restart mechanisms.