Sorry Ned, I was referring to the "cheaper" part claim. Your
recognition for a shorter idle residence time, cache time, was
previously noted.
I have yet to compare notes with customers. Among the top senders is
FaceBook's mailer with a 5 mins delay before sending QUIT. It looks
like Hoffman's imc.org mailer setup has a 4 mins delay and there are
others from unknown senders. So its apparent now where these odd
unexplained delays are now coming from - Connection Sharing clients.
Well, I can't stand for this. A local policy option to lower the
server idle client timeout to 1 minute (default) after the 1st
transaction has been finished will be provided. I'm calling it
Final-Idle-Time. Probably also look at some tuning logic to monitor
Sender vs Final-Hold-Time to see if that can be used to kick out idle
clients when new connections come in. This is part of the problem
that a CS logic does - increases cost on receivers to avoid the
potential problems caused by them, and that includes contributing to
any DoS attacks.
ned+ietf-smtp(_at_)mrochek(_dot_)com wrote:
ned+ietf-smtp(_at_)mrochek(_dot_)com wrote:
>> I don't think the practice of connection caching is particularly
selfish when
>> compared to the cost of having the connection torn down and then
re-established
>> with some frequency, when it's generally much cheaper for both the
sender and
>> the receiver to just leave it open.
>
> Exactly. Although it is necessarily up to the client to decide, the
server also
> benefits as long as the connection isn't cached for very long.
I can see this when two systems have prearranged dedicated channesl
for high load exchanges, but in general, I disagree that it is
"cheaper" and definitely not for the receiver. I don't see how that
can be justified. Average SMTP session times last year of 10-15
seconds are now up to 4-7 minutes this year Thats better?
Please reread my message. Nowhere did I say that clients should cache
connections up to the 5 minute timeout - in fact I said the exact opposite:
Caching a connection for anything even close to five minutes is almost
certainly counterproductive - SMTP connection establishment isn't *that*
difficult. But holding on to a connection for just a few seconds can be
quite
beneficial for systems handling high volumes of mail.
I have no idea why your servers are seeing longer averages, but I can
assure
you it isn't our client that's doing it. When there are no more messages we
only keep connections open for a few seconds and this is not configurable.
Ned
--
Sincerely
Hector Santos
http://www.santronics.com