Development of router clusters to provide fast failover in IP networks
P. L. Higginson, Mike Shand · 1997
this paper, we start by discussing our targets and the behavior of existing routing or bridging protocols and follow this with a detailed analysis of the different failure cases. We then show how we have modified the behavior of the routing control protocols to achieve the desired failover times on links between routers or in response to the failure of intermediate routers. Finally, we describe the new IP Standby Protocol and the mechanisms we developed to achieve fast recovery from failures on the LANs local to the end hosts. This part of the problem is the most challenging because the hosts are of many types and have IP implementations that cannot realistically be changed. Thus all changes have to be made in the routers. Our secondary aims were to allow the use of router clusters in any existing network configuration, not to constrain failover to simple pairs of routers, to be able to share traffic between available routers, and to continue to use the Internet Control Message Protocol (ICMP) redirect mechanism for optimum choice of router by hosts on a per destination basis. A common problem of hosts is that they do not time out redirects. This problem is avoided by the adoption mechanism within the router cluster. Having met these aims, as well as fast failover, we can justifiably call the result router clusters.