So, my colleague (half sys-admin, half perl guy) decided to investigate inetd for our little client-server app.
Not only does he have it up and running in less than a week, he's cut the actual Perl code down to about %20 of what it was before. The big advantage to this approach was error logging, but it has many other advantages as well: only runs when called, virtually no chance of die'ing short of a hardware failure, no need to hand create sockets and no need to specify the port.
The drawback to this approach is that it's more overhead. If the server is "off" (after 30 seconds of inactivity - also configurable) it must reload the server each time it's "activated". For lightweight servers this is no big deal and ours fits the bill.
The other drawback is that it requires root access. Since we don't have admin rights on most of the machines, we have to rely on a very unreliable sys admin to implement this, assuming he'll even agree to do it in the first place, the lazy bastard.
