Mike

I'm pretty sure you're experiencing a standard problem that most of us=20=

have been bitten by at one stage or another.  Ged is right when he says=20=

that 120 apache children at 40MB a pop will eat your RAM up.  The piece=20=

of the puzzle that you're missing is this:  due to the dopey VM code in=20=

Linux, when you run out of physical ram, processes get swapped out and=20=

at that instant, they become UNSHARED.  In other words, if your process=20=

is taking up 40MB, of which 30MB is shared, and it gets swapped out, it=20=

will get swapped out as a 40MB process.  When it gets swapped back in,=20=

it will take up 40MB of your precious RAM, and will probably force=20
another apache process to get swapped out and unshared, which will=20
cause another...

It's a tricky problem, but here's how you solve it:

You must, at ALL times, keep your machine from swapping out Apache=20
children.  This is tricky.  If you're running other things on your=20
machine, like a DB (I assume you're not since you say you're using=20
Oracle) then it becomes nearly impossible to prevent.  But if your=20
machine is pretty much just Linux + Apache, you're OK: just make sure=20
that

MaxClients * LargestPossibleChildSize + AllRamUsedByOSAndOtherProcesses=20=

< 4GB

Since you're running a front-end Apache too, you need to figure that=20
into the equation.  But the key here is that you can NEVER swap out an=20=

apache process, or you'll end up in the downward spiral of death, which=20=

seems to be what's happening to you.  There's no recovery from that=20
unless you can stop apache; only other solution is to reboot.  If=20
that's the case, you might as well be using Windows XP.

Hope this helps

Kyle Dawkins
Central Park Software


On Monday, Oct 13, 2003, at 05:34 US/Pacific, Mike Norton wrote:

>  We are currently experiencing some problems with our mod_perl setup=20
>  the server we are using is a Dual Xeon 2.4Ghz running Redhat AS 2.1,=20=

>  4GB Ram ,=A0Apache/1.3.27 (Unix) mod_perl/1.27. When the server is=20
>  under=A0a heavy=A0load it all of a sudden consumes a large amount of =
swap=20
>  and the server load shoots through the roof, the machine normally sits=20=

>  quite happily not even touching the swap normally however it seems to=20=

>  consume anything up to 3GB of swap all in one go. Does anyone have any=20=

>  ideas on how to resolve this issue or why it happens ?
>  =A0
>  Thanks
>  =A0
>  Mike