Currently the LiveJournal web servers are idle (CPU-wise) and the databases are fine (disk-wise). But the site is slow because web servers are all blocked trying to send outgoing email. (we have a tool to let us see why the site is blocked, and we're seeing mail taking on average 2.6 seconds)
Basically the problem is our mail architecture kinda sucks. (I imagine) Mostly I imagine this because we hardly touch it, we've hardly modified it from its defaults, and we don't really know what we're doing when it comes to running high volume mail systems.
Considering the mail servers don't look disk bound, I don't see why web servers should be tied up for 5-10 seconds some times sending mail, getting either a success or failure.
What I want to happen is for web servers to inject outgoing mail into Postfix via SMTP (which we do now), but then have Postfix write the message to disk and work on sending it later. I do *not* want the web client to wait several seconds while Postfix actually contacts its next hop and tries to deliver it.
Is that behavior configurable? I hope it's not already doing what I want, otherwise I'm confused.