In article <F03BA3F206076D16E3DF5BDE@PSB> you write:
My entirely subjective impression is that the typical message
content transmitted over SMTP is moderately long (in terms of
packets and acks). That differs from years ago when SMTP was
often used to transmit a lot of what amounted to control
messages ...
Agreed. The average number of recipients in legit mail has also
dropped to close to 1, because bulk mail is pretty much all customized
per recipient, so the number of round trips even without pipelining is
pretty small. (There's a Chinese spambot that sends 21 receipts per
message, but it doesn't count.) If you want to get more mail through,
send it in parallel. My not particularly large mail server easily
handles 50 connections in each direction and I could increase it to
200 if I needed to.
Also, checking for clients that send HELO/EHLO before the greeting
remains a surprisingly effective way to recognize spambots. You could
sort of imagine ways to whitelist clients who'd seen an early
pipelining flag, but in the common case that there's a bunch of MTAs
behind a single pool of IPs and a load balancer, keeping the cache in
sync and guessing what IPs are in a client pool wouldn't work very
well. I'd rather crank up the concurrency than lose early talker
checks.
R's,
John
_______________________________________________
ietf-smtp mailing list
ietf-smtp(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf-smtp