If you have a Web cluster with fifteen plus machines, are you really going to run an MTA on each one?

If I have a web cluster with fifteen machines, running 15,000 programs, I rather have to deal with 15 MTAs than 15,000 - because every program that delivers mail is an MTA.

Should each user in a company have an MTA on their local workstation just because.

Yes, but that's irrelevant.

Yes, networks can be unreliable, but then, if they're down, who's using your application anyway?

Just because you can't reach your mail server because of network problems doesn't mean you are unreachable by someone else.

Secondly, if your MTA is down, don't you think that's going to get fixed right away?

Sure. But what if that's going to take a few hours? Even in a company where there's 7x24x365 support, it might take a while before the problem gets fixed.

And yes, services can be overloaded, but what does this mean?

It could mean the remote service is not accepting connections. What should your program do, drop the mail in the bitbucket? I rather have it queued and retried, it could be important.

I'd rather not have my application lose it's mind just because I kill off the local MTA, or change it to something more secure than sendmail.

As long as the replacement comes with a drop-in replacement for sendmail, there isn't a problem. I know qmail does.

Abigail