On Thu, 27 Sep 2001, Sheldon Hearn wrote:
> I'd imagine that routing half a million single-recipient messages
> "one by one" take a long time indeed! Will multiple -qq processes
> interoperate correctly if I launch a few myself? I'm guessing no. :-(
You are right, because each one will work its way through the entire
queue, so there's no point.
> Nope. Even if I could tell it to stop after routing, having a single
> process performing all routing is just too slow. I need parallelism
> in routing. Right now, I achieve this by sacrificing more efficient,
> hint-based delivery.
You can get parallelism in routing by using -odqs (aka queue_smtp_domains),
but you also want to inject mail faster than it can be routed. *That* is
the problem.
How about modifying your injection process? Use queue_smtp_domains (or
-odqs), and during the injection, have a look at how many exim processes
are running. If the number is greater than n, hold up the injection
process until the number drops below n. This will slow down the
injection rate, but do the routing while it is happening. I imagine you
could set n to several hundred, possibly more.
The result of this would be a humungous hints database listing all the
messages for the various hosts. How that would perform during delivery,
I wouldn't like to speculate. But it would be interesting to try!
--
Philip Hazel University of Cambridge Computing Service,
ph10@??? Cambridge, England. Phone: +44 1223 334714.
