Information Technology Reference
In-Depth Information
Identifying the Starting Router
The starting router is the first router to be migrated to the new architecture. There are three
choices; each has its own challenges:
•
A core router
—If a core router is migrated first, traffic can be black-holed during the
migration. You can avoid the temporary traffic loss by moving the core router out of
the forwarding path if redundant connections exist.
•
An access router
—If an access router is migrated first, the reachability of cus-
tomer prefixes may be affected. If customer prefixes are originated from that router,
these prefixes are unavailable unless they can be generated in other routers tempo-
rarily. If customer prefixes are exchanged via BGP, these prefixes are unavailable
unless the customer is multihomed. In this chapter, access routers do not have eBGP
sessions with customers.
•
A router with external peerings
—If an eBGP router is migrated first, connections
to the external neighbors are affected unless multiple connections are available.
In the topologies presented in this chapter, redundant connections exist between the core
and access routers, and each POP has at least one core router without a single-homed
external connection. Thus, a core router is the most logical choice as the starting router
for all case studies.
Minimizing Traffic Loss
During migration, traffic can be temporarily looped or black-holed. Traffic loops are gen-
erally caused by conflicting next-hop settings between routers. Two types of next hops are
involved: BGP next hop and IGP next hop. BGP next hop is often recursively resolved to
an IGP next hop. The case studies in this chapter discuss the cause of potential loops and
ways to prevent them.
Traffic loss caused by
black-holing
occurs because a router on the forwarding path does not
have the correct routing information. A router under migration drops all traffic for which it
has not yet learned routing information. The solution to the problem is to move the router
temporarily out of the forwarding path for all prefixes or to build additional BGP sessions
in advance. Specifically, you can choose from the following options:
•
Shut down all relevant links on the router under migration. This approach is simple to
accomplish but might require remote console access to the router to ensure further
migration. The other drawback of this method is that BGP sessions might not be able
to form while some of the links are down.
•
Change the IGP interface metrics so that the router under migration is not selected as
an IGP next hop. While the router does not have transit traffic to forward, its BGP
sessions can be established; thus, routing information can be exchanged. Undoing the
metric changes puts the router back in the forwarding paths.
