The BGP table contains hundreds of thousands of routes. Which design technique improves the convergence time?

When a BGP routing table contains hundreds of thousands of routes, the resulting UPDATE messages can exceed the path MTU and be fragmented at the IP layer. TCP path MTU discovery allows the BGP TCP session to dynamically learn the optimal MSS along the path, avoiding fragmentation and the associated retransmission overhead, which directly improves the speed at which large routing tables are exchanged and convergence is reached.

BGP is standardized to use TCP destination port 179; using non-standard ports does not affect UPDATE message size, fragmentation, or the efficiency of routing table exchange.

Route reflectors reduce the number of required iBGP sessions from O(n^2) to a hub-and-spoke model, which improves scalability and session overhead, but they do not address TCP-level fragmentation that slows the exchange of large routing tables.

BGP confederations divide an AS into sub-ASes to reduce full-mesh iBGP requirements and improve scalability, but like route reflectors they do not resolve the IP fragmentation issue that impedes convergence when transmitting hundreds of thousands of routes per session.