Ruediger Pluem wrote: > > I am not sure if doing this for each connection (not each *request*) > is really that much of a performace hit. I used a modified version of > mod_limitipcon and do just that. So far I haven't noticed any performance > issues with this approach. But maybe with a maximum of 1040 clients > I am just having a small server :-).
The problem is which pain do you want; single-connection contention over a lock to a IP resource counter which may bottleneck, or multiple thread or cpu examination of the scoreboard. Presuming the score never swaps, and is frequently queried, I'm speculating that this approach is more robust. As far as niq's "already too late" comment, if N+x requests will sink the server, where x is the number of CPU's which may be constructing new scoreboard slots, then N was already set too high.
