On 30/10/13 09:55, Nikita Koshikov wrote:
> This is my acl for relaing hosts:
> accept hosts = +relay_from_hosts
> control = submission
> control = dkim_disable_verify
>
> +relay_from_hosts - is just ip-based list.
>
> Yesterday I have added new control - cutthrough_delivery and this is my
> results:
>
> Every morning I have script that sends mail to ~570 addresses.
I assume this is many one-recipient messages, given sequentially to Exim?
> And before cuttthrough it takes 2 minutes to accept and delivery all this
> messages to recipients.
> Today (with control = cutthrough_delivery ) I saw in logs new ' >> ', that
> confirmed it's begin to work but the last message was delivered after 16
> minutes from start time (5:55).
>
> And my question - is it expected or I made something wrong ? I thought that
> new option should speedup delivering process.
I suspect that without cutthrough you end up queueing and then
delivering in parallel, whereas with cutthrough the sequential
presentation to Exim is maintained all the way through to (many of)
the deliveries, resulting in higher wallclock time.
If your dump of messages could be given to Exim in parallel you'd see
different - but arranging this probably isn't worthwhile. If they
all come down one smtp connection you could, perhaps, only
decide to use cutthrough on the first?
Cutthrough does not necessarily give a speedup.
--
Cheers,
Jeremy
