> the reason I suspected the net tests was because the problem happened after > we upgraded from version 2. we were keeping up before. I was thinking it > was the net tests because I wasn't doing them in 2. I wasn't aware of other > changes from 2 to 3 that could be the cause. now if two used fewer > children, that could indeed be a smoking gun.
Most people that have seen a slowdown in 3.x seem to be due to thrashing due to larger memory usage. This was especially bad on 3.0 and 3.0.1. Going to 3.0.2, cutting down the number of children, and cutting down the number of messages each child processes before restarting are the general solutions. That said, some people have seen problems with slow net tests. But that seems to always turn out to be a local problem that isn't specifically SA once it is analyzed. Loren