Actually, killing your service when it's overloaded should prove to be a good load balancer over the long term. Initially, the server will be killed a lot, preventing users from using your service. At some point, those users won't come back anymore because your service is so bad (unless it's Twitter), and you won't need to kill the server anymore. Eventually, you won't have any users, so you won't have any load problems.

I will stick with Perlbal, however.

Ironically, the original article's server appears to be down.

Maybe there is a god out there.

Or maybe humans just have a tendency to seek and recognize logical patterns in random occurences and then confirmation bias does the rest.

I doubt swombat was being serious. He was merely implying that he was amused by this random occurrence.

No, really, I totally took the fact that the server was down as proof that there is a god, and the bible is correct, and Sarah Palin should be the next president.

Seriously.

If you're using round robin dns, killing the web server will force clients with that particular server cached first to connect to a different one.

Not particularly elegant, but not a complete fail.

kills -> restarts? Http can be pretty stateless anyway.