Figure 21 shows a sample topology where source end system A is communicating with destination end system B through routers 1, 2, 3, and 4.
Figure 21: Transit Router Topology
The transit routers, 2 and 3, learn the route to B from BGP. In a steady state environment, the BGP routing tables are synchronized on all the transit routers.
Suppose the traffic forwarding path is currently A –> 1 –> 2 –> 4 –> B. If transit router 2 goes down, the network converges to the alternative path, A –> 1 –> 3 –> 4 –> B. Because transit router 3 already had synchronized its BGP routing tables, traffic forwarding continues without delay.
When transit router 2 reloads, it establishes adjacencies with routers 1 and 4, and sends out its LSP advertising its neighbors. While router 2 begins to synchronize its BGP routes, the network reconverges to the original path of A –> 1 –> 2 –> 4 –> B. Traffic from A to B is forwarded to router 2. Typically, BGP has not converged by then, so router 2 does not have the BGP route that it needs to forward the traffic, and drops the packets, resulting in a black hole until the BGP convergence is complete.
You can avoid this black hole by configuring the overload bit for the transit router. In this circumstance, router 2 sends out its LSP with the overload bit set in its header as soon as it reloads, before it establishes all adjacencies. The bit set in the header indicates to all the routers in the domain that router 2 is overloaded and not to use it to carry transit traffic. The forwarding path continues to be the alternative path, A –> 1 –> 3 –> 4 –> B, even after router 2 reloads.
When BGP convergence is complete at router 2, router 2 sends out a new LSP with the overload bit cleared. The other routers then include router 2 in their SPF calculations and revert to the original path, of A –> 1 –> 2 –> 4 –> B.
 | Copyright © 2008, Juniper Networks, Inc. All rights reserved. Trademark Notice. | |