Nikolai Dahlem:
> Am 2018-03-09 13:13, schrieb Wietse Venema:
> >> delay=0.51, delays=0.21/0/0.17/0.13
> > 
> > Just to be sure, these numbers include receiving and delivering mail.
> > 
> > 0.21        time from message arrival to last active queue entry
> > 0           time from last active queue entry to connection setup

For comparison, inbound mail from the postfix-users list to my own
machine typically has "delays=0.13/0/...". Ping from my machine
to the list server says ~5ms.

> > The first number shows that it takes 0.2s to put mail into the
> > queue. Does your host run a LOCAL caching DNS server? If not, that
> > could explain sluggish performance for receiving and sending mail.
> > 
> > When receiving SMTP mail, Postfix looks up IPaddress->name and
> > name->IPaddress which can result in two DNS queries per inbound
> > connection.
> I run a local instance of unbound for dns caching

I also notice that you have configured a Milter.

To determine whether it is IPaddress/name lookup please re-run the
smtp-source experiment and set the -d flag to keep the SMTP connection

    time smtp-source -d -4 -m 20 -S Test -l 500 -t u...@xxx.yy

To determine whether it is Milter or something else, please re-run
the smtp-source experiment after

        $ postconf -# smtpd_milters
        $ postfix reload

smtp-source is talking to a local address, therefore network latency
should be negligible UNLESS you turned on traffic shaping in the

> > The other numbers show that the total delays are dominated by
> > deliveries. Which Postfix delivery agent that was producing these
> > numbers? I suppose that was the SMTP client? If it is a local
> > delivery agent, then WHAT IS WRONG WITH YOUR FILE SYSTEM?
> It was the SMTP Client.

As for outbound performance, that depends on network distance. 

 0.17        time in connection setup, including DNS, EHLO and STARTTLS
 0.13        time in message transmission (including response to '.')

Was this mail delivered over TLS? How large was the message?
Did the remote server announce PIPELINING in the EHLO response?

All these questions are better answered with a tcpdump recording.


Reply via email to