Author: Peter D. Gray Date: To: exim-users Subject: Re: [exim] default value seems a big low
On Thu, Aug 17, 2006 at 10:34:59AM +0100, Philip Hazel wrote: > On Thu, 17 Aug 2006, Peter D. Gray wrote:
>
> > After 10 messages are injected via SMTP on the same
> > connection, further messages are placed in the queue
> > rather than being handled immediately.
>
> <snip>
>
> > Can someone explain the rationale behind such a low value?
>
> The rationale for having a limit is that otherwise Exim starts a
> delivery process for each message as soon as it arrives, and if 1000
> arrive quickly you would suddenly have 1000 delivery processes all
> trying to run at once.
>
Yes, the limit is a good idea.
> > It would seem to me that treating messages delivered in the
> > same connection differently to those arriving
> > on different connections is not actually that helpful
> > but maybe I just have not thought this through enough.
>
> You can limit the number of different connections. If you have set that
> to, say, 100, and messages come in very fast over those 100, you might
> have (with the default setting) 1000 deliveries going on simultaneously,
> but no more. Without the limit on each individual connection the number
> could get a lot higher. Agreed, it's unlikely that all 100 connections
> will be spewing messages in this fast, but some limit is needed IMHO.
>
> As to the value of 10, well, yes, it's small. Partly that was me being
> conservative, and partly it's because it was invented some time ago
> (1998) when machines were less powerful, and partly it's because most
> connections only send one message anyway...
>
I think you should at least consider making the limit 0 by default.
It seems to me that most (in fact nearly all)
of the "rate limiting" settings
in exim are turned off (eg queue_only_load). The assumption
seems to be that if you want/need rate limiting, you turn it
on yourself. The default value for smtp_accept_queue_per_connection
seems a notable exception and may catch people out. It certainly
caught me. As MTA's get smarter, they tend to batch things
more and more.
