At the web hosting/devopement company I work for, we have a client that has a large mailing list (over 250000, opt-in, yes). When I came on board the situation was that sending the list constantly killed Sendmail and it took nearly six days to send. Some yahoo put a 2 second delay between each email. Of course, they also used perl to make a system call to /var/lib/sendmail. Yuck. That's process/filehandle/cpu hell.
The solution: P166 with Debain 2.2r2 running Qmail and a change to the way the mail server was used. Instead a making a new connection for each message, one connection (via Net:SMTP) was created and used until the list was done. The sending machine used almost no cpu time (~2-3 %) and hardly any memory (the entire list is never in memory at once.
The mail machine queue can get as high as 100000, which is insane, but qmail doesn't care. Incoming connections never slow down (that I've noticed, anyway) and surprisingly still seems to actually deliver almost 1 message per second (T1).
The entire list takes less than two days to finish being delivered. The only problem is that the P166 isn't beefy enough to start qmail-remote processes fast enough. :( (I've set the max at 75, which it hits easily once the incoming onslaught stops and just the queue needs to go out)
As others have said, it's the job of your MTA to figure out the best way to send the queue. Don't bother with that (IMO) but rather get a POS machine that no one else wants to use and turn it into a bulkmailer.
gralbe
gralbe@thebelliconnection.com
P.S. Feel free to mail me if you have any questions or think I'm an idoit.
| 