On 12/04/06 14:51 -0700, David Lang wrote:

> but if you can accept latencies of hours then you have a lot of leeway to
> spread your peaks out and your peak bandwidth can be far closer to your
> average bandwidth, so your statement that the peak is 20-30x the average
> doesn't sound reasonable.
> what am I missing here?

The peak load is generated by the injection of a message. This can take
the form of a single message to multiple recipients, or one message per
recipient. If marketing purposes are involved, the second is the norm.

When you smack a few thousand messages into a queue, the MTA tries to
push them all out simultaneously. This creates a choke, causing minutes
worth of latency. Remote MTAs may be down, or they may have greylisting
enabled, in which case the delay increases even more. Mail flow
typically looks like (actually it looks worse, but those details can be
ignored for the moment)
______ <----- Network saturated.
| \
________| \______________/\____________________

Typically, you do not want the second smaller peak at all, but that can
be lived with. More mail clogging the queue is a bad idea. For a good
list, something like 99% will go through in the first blast.

The hours of delay are exceptional and preferably avoided.

Devdas Bhagat
firewall-wizards mailing list