On Sat, Jul 22, 2000 at 12:45:57PM -0700, [EMAIL PROTECTED] wrote:
> I've written a little perl script to analyze a qmail log.
Have you looked at qmailanalog? Could it help you if it does not
already do what you want?
> This scripts gives a hint as to what you might save in bandwidth
> if qmail supported multiple recipients.
The zoverall script in qmailanalog will give you a maximum bound to this
number. On my SMTP proxy (450MB over 9.5 days, not that big yet), a
maximum of 20% could have been saved.
> This results is indicative at best - here are some caveats:
>
> o failed deliveries are not counted
Reasonable, since nearly all failed deliveries will fail before the
"DATA" command.
> o Aggregation is by FQDN, not MX target
Which is the only reasonable way to do it. If you aggregate based on MX
target, you need to do (and wait for!) DNS lookups on all recipients of
each message. This is a good way of slowing things down for no real
gain.
> o The incremental costs of subsequent deliveries via multiple recipients
> is assumed to be zero.
Which is one of the contentious points in the whole discussion. This
one *REALLY* needs some real-world measurements, which would be quite
difficult to do. There will likely be a point (in terms of message
size) where the time cost of opening up more connections (in parallel,
remember) will be less than the cost of issuing another RCPT.
You could simulate this by producing a test message, and (1) forking off
N copies of qmail-remote with a single recipient, and (2) forking off 1
copy of qmail-remote with N recipients, and time how long it takes for
the qmail-remotes to exit. Repeat with a series of message sizes. On
my proxy again, the median size is around 3000 bytes (including
headers), just as a guide for how to distribute the sizes. Make sure
the system you benchmark with is far enough remote to cause significant
latencies (100ms or worse), or try various systems with various
latencies.
> Since the script is only lightly tested, I'm soliciting a few volunteers
> who are willing to run this script on their log files and send the results
> back to me (and/or the list if you so desire).
I'd be willing to do this, I'm somewhat curious myself.
--
Bruce Guenter <[EMAIL PROTECTED]> http://em.ca/~bruceg/
PGP signature
