> As such, if iBGP by TCP mesh without reflectors is not acceptable, Says who ?
See RRs were deployed to address three points: #1 - Configuration simplicity #2 - Path reduction #3 - TCP session reduction So: #1 is a real operational problem and we do have drafts which add BGP Autodiscovery. As example: https://datatracker.ietf.org/doc/html/draft-raszuk-idr-bgp-auto-discovery-08 New work could have built in auto discover day one. #2 No longer a reason since ADD-PATHs ALL used often nowadays results in exactly the same amount of BGP state. #3 In the old days keeping 1000 of TCP connections in custom kernels were an issue. These days TCP implementations on linux go orders of magnitude higher. Then there is QUIC ... So practically if #1 is addressed full mesh of IBGP can be easily deployed today. Thx, R.
