On Sat, Sep 08, 2012 at 10:49:49PM -0500, Stan Hoeppner wrote:
> Is the latency you describe caused by disk IOPS or Postfix queue
> processing behavior, or ?? I'd assume most folks using a 2nd instance
> for fallback relay duty have the fallback instance queue on the same
> physical disks, so I also assume the latency isn't IOPS related.
It is caused by either or both of the following:
- Too many deferred messages in the active queue starving
processing of the incoming queue.
- Exhaustion of delivery agent process limits underserving
good destinations while being throttled by bad destinations.
It is unlikely that disk I/O or network I/O are at issue with busy
queues that contain lots of mail to high-friction destinations.
Output processing is not particularly disk intensive (that cost is
paid on the input side). The network is rarely a factor, SMTP is
not bandwidth intensive, or if it is, once the pipe is full, one
is delivering a lot of mail about as quickly as possible, so that's
fine.
--
Viktor.
