[EMAIL PROTECTED] wrote:
>On Thu, Mar 30, 2000 at 03:23:21PM -0500, Dave Sill wrote:
>> Logging with multilog is very cheap.
>
>sounds interesting... where do I look?
ftp://koobera.math.uic.edu/www/daemontools.html
>:-) Just an idea on how to split the mail:
>
>I guess that "a large percentage" of said deferred mail goes to a
>handful of domains. So I would take that spare machine and set it
>up to handle such mail, and configure the main mail server to
>forward that mail to the second mail server using "smtproutes".
>So you would deliver most of that mail immediately to the
>secondary server, thus freeing the main server's resources,
>and have the secondary mail server trying the slower deliveries.
That's an interesting approach. I'm still unclear about the advantages
of splitting the load between servers by deliverability vs. some
arbitrary scheme. How does having the main server pumping out all
highly deliverable mail and the fallback server plodding away on
frequently deferred mail win over two (or three or N) servers doing a
mix of both?
>The remaining problem is to have a current set of smtproutes
>available which could probably be generated from log file
>analysis in near-realtime.
Doing it anywhere near realtime would require many qmail restarts on
both systems, which is not good for performance.
-Dave
