Conclusions Drawn from Evaluating Fault Detection and Recovery Times

We presented several approaches for increased availability for network designs by evaluating fault detection and recovery times and the adverse impact on computing and memory resources. In comparing these networking designs, we drew the following conclusions:

• Link aggregation is suitable for increasing the bandwidth capacity and availability on point-to-point links only.

• Layer 2 availability designs using Sun Trunking 1.3 and Split MultiLink Trunking available on Nortel Networks Passport 8600 Switches were configured and tested. Distributed MultiLink Trunking was also configured and tested using Nortel's smaller Layer 2 Business Policy switches. Both switches were found to provide rapid failure detection and failover recovery within two to five seconds. Further benefit of this approach was that the failure and recovery events were transparent to the IP layer.

• Spanning Tree Protocol is not suitable because failure detection and recovery are slow. A recent improvement, IEEE 802.3w Rapid Spanning Tree, designed to improve these limitations, might be worth considering in the future.

• Layer 3 availability designs using VRRP and IPMP offer an alternative availability strategy combination for server-to-network connection. This approach provides rapid failure detection and recovery and is economically feasible when considering the increased MTBF calculations. Be sure to investigate the processing capabilities of the control processor and consult with the vendor on the impact of additional load due to the ICMP ping commands caused by IPMP.