There's a problem in the server software. When the load gets high, it fails catastrophically instead of gradually. Robert and Patrick Collison are investigating, but they're still not sure what the problem is. My guess from the external evidence is that it's related to garbage collection.

Killing the server process fixes the problem, at least for a day or two.

Interesting. Are they investigating by load testing a dev environment, or some other method?

Patrick managed to reproduce the problem by writing something to replay server logs, but he's not sure yet what's causing it.

Interesting.

I'm assuming it's traffic-related, because I've run had ASV running on my server for weeks on end without real issue (I don't get much in the way of traffic).

Maybe there's something taking superlinear time.

Out of curiosity, how high is high?

Are you guys running your own web server? If so why? It would seem to me that apache wouldn't have garbage collection problems.

Its always garbage collection.

Are you running MzScheme on OS X? Maybe it's an issue with the background threads... If it's FreeBSD then scratch that.

FreeBSD.