Information Technology Reference
In-Depth Information
The MSS value is set to the MTU minus the 40 bytes of IP and TCP overhead. The 40-byte
values assumes that additional TCP options are not being used, which is the default behavior.
This provides the 1460-byte MTU. It is possible to have even larger MTU sizes, especially
internal to the network. The Packet over SONET (POS) link has an MTU of 4470. If two
BGP peers use PMTUD and are connected only by POS or ATM links with MTUs of 4470,
the MSS could be as large as 4430, which provides an even greater reduction in update
packets and TCP ACK messages.
There is a major caveat to be aware of when the MSS derived from PMTUD is greater than
1460. Figure 3-2 shows the initial network topology, with the path of the BGP TCP session.
Figure 3-2
Path MTU Discovery Path Change Scenario
R2
MTU 4470
MTU 4470
R1
R3
MTU 4470
MTU 4470
MTU 1500
R4
R5
Initial Path
Failure Scenaro Path
If the link between R2 and R3 fails, the TCP session is rerouted and sent over the link
between R3 and R5. This path has a Fast Ethernet segment, which reduces the path MTU
to 1500. The result is that large BGP updates are fragmented on R5, which reduces perfor-
mance and increases convergence time.
However, the impact that BGP update fragmentation has depends heavily on network events.
If no major routing changes happen, there is minimal to no impact. If major routing churn
occurs, this can significantly impair convergence, especially if the link failure results in a
large number of BGP sessions experiencing this impact.
This issue is resolved by using Gigabit Ethernet and jumbo frames. It is also becoming
much less common to see Fast Ethernet as part of the core infrastructure in large BGP
environments. A typical large-scale BGP network is composed of POS and Gigabit
Ethernet.
 
Search WWH ::




Custom Search