Hello, > > way we can find the location of the problem. I have not been able to > > reproduce the problem here, by the way. It's probably some subtle bug > > in the code.
Just a note on reproducing it; if the load on our servers stays low, the problem isn't detectable. I think the pop3 daemon is still using quite high cpu time, but it's fairly short lived. When we get numerous long-running connections (maybe 4-6 high-speed users pulling a lot of pop3 data) and then throw the other ambient connections on top of that, things seem to degrade, I think just in cpu contention. The processes that stay running long enough to trace them don't appear abnormal other than the write() calls occur fairly slowly, compared both to what you would anticipate a high-speed connection to be capable of and compared to watching the same processes at other times when the problem is not happening (ie. not having as large of an impact). -- Jesse Norell jesse (at) kci.net
