Information Technology Reference
In-Depth Information
The decision process on the router reflectors could result in prefixes being withdrawn
and updated across the entire network for a subset of those 80,000 prefixes received
from the peering router. The peering router also runs the decision process and might
withdraw prefixes from the route reflectors based on the decision process outcome.
This BGP scenario results in the peering router's receiving 80,000 paths from the peers
and 250,000 paths from the upstream route reflectors (125,000 from each). The route
reflectors each receive 80,000 prefixes. If 25 percent of these are installed in the
routing table on the route reflectors, this results in 20,000 update messages being sent
to all iBGP peers and route reflector clients on those route reflectors. A total of 50
peering sessions on the route reflectors would result in 1 million advertised paths per
route reflector.
•
Scenario 3: The route reflector initializes
—A route reflector initializes its BGP ses-
sions with its regular iBGP peers and route reflector clients. The route reflector might
receive 400,000 paths, which results in 125,000 prefixes installed in the routing table
from its nonclient iBGP peers. The 125,000 prefixes are advertised to all its client ses-
sions. If the route reflector has 50 clients, 7.5 million prefixes are advertised. If the
route reflector has 100 clients, there are 15 million prefix advertisements. This provides
a best-case number. If the sessions come up staggered over a period of time, the best path
could change several times, resulting in a significant increase in advertisements.
Each of these scenarios results in different behavior on the network. The concentration of
update generation varies based on the source of prefix information for that BGP speaker.
The edge router scenario injects very small amounts of new prefix information into the net-
work, resulting in a much smaller scope of impact. The peering router injects significantly
more new prefix information, which greatly increases the scope of impact on the network.
The route reflectors, however, are the real workhorses for distributing BGP prefix informa-
tion. The route reflectors send the most significant number of prefix updates, regardless of
where the route initialization takes place.
The commonality in the initialization scenarios is that BGP works with large amounts of
prefix information. Changes in that information can result in very large amounts of data
generation and advertisement. This section focuses on tuning BGP to deliver large amounts
of information as efficiently as possible. The areas in which tuning is performed are as
follows:
•
TCP operation
—The underlying transport for BGP is TCP. The operation of TCP
provides opportunities for improving BGP convergence.
•
Router queues
—The route reflectors generate a moderate amount of data per peer,
but aggregate for all the peers results in a significant amount of data generation. The
operation of TCP requires that the peers respond to the route reflectors to acknowledge
receipt of the TCP packets. This many-to-one data flow can overload the route reflector
with more information than it can process immediately.
