Hi All,
Wondering if i can find a definitive answer from the great people here.

I've been trying to set-up/tweek these param's in Apache to a setting suitable to the server for the amount of memory. When i look around some people say, hey.. just look at the memory used per process and then divide that by the amount of memory available and you get the number of processes that can be handled by Apache in one go before it starts swapping.

Well, for this i'd done this calculation and for me it turned out to be 200 approx concurrent connections.

Well funny thing is, out MySQL server had a slow down so the Apache servers were running at approx 450 concurrent connections and weren't swapping in memory etc (still had 600MB free not including what was available in Buffers/Cache - 'free -m')...

Thus if i had have set the limits to 200 then people would have been not getting to the site, so i'm kind of pleased that i happened to not have the time to set this yet.

so the question is, although the answer seems obvious, is this calculation correct?
Cheers All,
M

I don't have atm any apache in production env., meaning that the only apache i'm running is for the developers.

But a few weeks ago i found this article Staying Out of Deep Water: Performance Testing Using HTTPD-Test's Flood

Perhaps you can run some stress tests to your apache with this tool and post a feedback.

Hope this help.

That is an interesting idea, I probably will do just that.

Is there anyone here that is able to once and for all say if this calculation is correct? In my mind it isn't, and i'm trying to work out to a very close proximity just how much each of out web servers can cope with, so i can work out scalability requirements.

Regards,
M

Another strange thing with this is that i have my apache servers monitored on all my boxes. So i could see during this slow down in mysql there were circa 400 simultaneous requests in each server (normally hovering ~150). The thing is using 'ps -ylC httpd --sort:rss' i can see the variance of each apache process memory usage, and in prefork mode this would mean approx 400 simultaneous child processes?? I'm pretty sure thats dead on correct...

Thus with each apache process generally taking up 15-24MB according to the output of the command above and looking at the RSS column (FYI top shows similar as expected!), then simple calculation 15MB x 400 = 6000MB memory usage supposedly???

AND yet no swapping was occuring at the time and still had 600 odd MB free mem on a box with only 4GB which a bit of would obviously have been in use on for the OS etc.

In which case, is anyone able to explain the theory that seem touted all over the place of this calculation? It must clearly be incorrect?
Thanks all,
M

PS - What about if i just set startservers at to some ridiculous high number like 700, and see if thats enough to make apache have too much memory and require swapping? And doing this whilst the server isn't even live? Same amount of memory is taken by each new process anyway right? Regardless of if its idle, so surely instead of running a set of load tests.... this method would surely work as well??