On Tue, Sep 01, 2020 at 02:30:05PM +0200, Stephan Seitz wrote:
> I try to understand different mail delivery times.
Sure, that's a quantitative analysis, and doing requires looking in
detail at the delivery latencies in the logs of both client and server.
To get help with this, you'll need to post detailed quantitative
metrics, anything else is just pointless wild speculation.
Ideally, after looking at the logs, but before you post them, just
the act of collecting them will help you realise what the problem is,
assuming there is a problem. You'll to look at both clientside and
server side logs (DO NOT enable *verbose* logging in Postfix).
And make sure that systemd-journald is not throwing away most of
your log messages, its default configuration is abysmal.
> I have a script (phpmailer) that sends 500 mails to a remote mail server.
Can the script log SMTP activity with millisecond or better time
granularity? Have you done that?
> The target address is a local /dev/null mailbox. The script uses
> SMTPKeepAlive = true; to keep the connection open.
Have you verified that the connection is actually kept open?
> If my client IP is part of mynetworks the time for 500 mails is about 29
> seconds.
This is anecdata, not data.
> If I use AUTH (SASL via Dovecot) the time is about 47 seconds. It doesn’t
> matter if the connection is protected with STARTTLS or not. The
> encryption isn’t the problem.
This is anecdata, not data.
> So I’m wondering why I have such a big time difference.
So are we.
--
Viktor.
