Doug MacEachern <[EMAIL PROTECTED]> writes:
> On Mon, 25 Sep 2000, Stas Bekman wrote:
>
> > All you care about is to measure the time between email sending start and
> > end (when the process continues on its execution flow). Why should one
> > care about the details of internal implementation.
>
> i only skimmed the first part of this thread, but assumed if you're
> talking about performance, you'd want to compare the overall impact on
> your system(s) of Net::SMTP vs. |qmail-inject. you cannot measure the
> overall impact just using Benchmark.pm is all i'm trying to clarify.
A better plan for such systems is to have a queue in your database for
parameters for e-mails to send. Insert a record in the database and let your
web server continue processing.
Have a separate process possibly on a separate machine or possibly on multiple
machines do selects from that queue and deliver mail. I think the fastest way
is over a single SMTP connection to the mail relay rather than forking a
process to inject the mail.
This keeps the very variable -- even on your own systems -- mail latency
completely out of the critical path for web server requests. Which is really
the key measure that dictates the requests/s you can serve.
--
greg
