Once upon a long time ago I was given the task of streamlining the processing of sending mail to a quarter of a million subscribers. (I was the senior sysadmin at a little anti-virus company and this was the list that people subscribed to for notification when the regular "defs" (virus pattern definition) files were posted to our BBS (remember those?), FTP and web sites).
My first step was to normalize that addresses (down to just [email protected]) and then sort that through the command: rev | sort | rev ... producing a list sorted by domains (and even sorted by subdomains with any upper level domain. This reduced queue processing time by about one order of magnitude (using a contemporary copy of plain old sendmail (The trick here is that most SMTP servers used to allow message pipelining --- allowing multiple messages/recipients to be processed with fewer separate TCP sessions).
Of course I also scaled it out over a couple of machines, later.
Send plain-text notifications only, keep em short and sweet. (Think tweet)
Use QOS on router/firewall to lower the priority of the outbound email.
Send in batches, say, 50 every 10 minutes.
Once upon a long time ago I was given the task of streamlining the processing of sending mail to a quarter of a million subscribers. (I was the senior sysadmin at a little anti-virus company and this was the list that people subscribed to for notification when the regular "defs" (virus pattern definition) files were posted to our BBS (remember those?), FTP and web sites).
My first step was to normalize that addresses (down to just [email protected]) and then sort that through the command:
rev | sort | rev... producing a list sorted by domains (and even sorted by subdomains with any upper level domain. This reduced queue processing time by about one order of magnitude (using a contemporary copy of plain oldsendmail(The trick here is that most SMTP servers used to allow message pipelining --- allowing multiple messages/recipients to be processed with fewer separate TCP sessions).Of course I also scaled it out over a couple of machines, later.