The issue here is that when you combine site overloading via SNI with CGNAT becoming much more prevalent on the client side, particularly on mobile networks, you wind up with a larger number of TCP sessions concentrated onto a few number of source/dst IPs, which means it’s more likely that you can hit the limit of 65K sessions between two given IP addresses (I won’t repeat the math that was shown earlier in the thread, but it looks solid to me). Most content providers really want to avoid this. As such, most website operators explicitly want to avoid sharing IPs with other sites.
-C > On Mar 2, 2019, at 2:55 PM, Ronald F. Guilmette <[email protected]> > wrote: > >> As for NAT and even web hosting, the 64k port limitation is also an issue >> as pointed out by others. > > No, it isn't. A web server needs one port (80). A mail server needs one > port (25). A name server needs one port (53). A /24 block provides nearly > seventeen *million* IPv4 ports for outbound _client_ use, most or all of > which should actually be migrated over to IPv6 anyway.
_______________________________________________ ARIN-PPML You are receiving this message because you are subscribed to the ARIN Public Policy Mailing List ([email protected]). Unsubscribe or manage your mailing list subscription at: https://lists.arin.net/mailman/listinfo/arin-ppml Please contact [email protected] if you experience any issues.
