On Sat, Jun 12, 2010 at 1:19 AM, Sean Conner <spc@xxxxxxxxxx> wrote: > It was thus said that the Great Bram Mertens once stated: >> Hi, >> >> http://httpd.apache.org/docs/2.0/misc/perf-tuning.html starts by >> saying "The single biggest hardware issue affecting webserver >> performance is RAM." >> >> However mod_status does not show information on memory consumption >> while it does display information on CPU usage. >> >> This seems odd to me. I have not been able to find any discussion on >> this in the archives so far so I'm hoping someone can shed a light on >> this. > > It's not easy. For example, one instance of Apache on my server (running > Lniux) is showing it's using 11M of memory (output of "ps aux", which > includes memory usage of each process), of which only 6M is actually in it's > "resident set size", which according to the documentation for "ps": > > resident set size, the non-swapped physical memory that a task has > used. > > But of that 11M, a good portion of it is shared between *all* the > instances of Apache---the actual code portion is shared between the Apache > processes, and any data that actually hasn't been changed since a given > child process started is at least shared with the parent (or main) process > [1]. I'm not aware of any easy way to determine the amount of non-shared > writable memory in a process. That would be my guess. > > -spc > > [1] Linux, as well as many other Unix operating systems, use a concept > of "copy-on-write." Memory is broken up into what are called > "pages"---typically around 4k each and each 4k chunk is made visible > (or "mapped") to a process. When a process creates a child process, > all the memory pages of the parent are visible to the child, but > they're marked "read-only" at this point. Reguardless of who writes > some data, the operating system can detect this (since the page is > "read-only" and all writes are trapped), that page is then > duplicated, so there are now two pages in memory with the same > content. One page is mapped into the parent and made read-write, > and the other one is mapped into the child and made read-write [2]. > Now each process has their own private copy of just that one page. > > It wouldn't surprise me if Windows works simularly, but since I > don't use Windows, I can't say for sure. > > [2] It doesn't matter which is mapped since the instruction that did the > write will be re-executed anyway. Thanks for the information. From this I gather I'm more interested in the overall memory usage of the server than that of apache itself. So I'll use those numbers. Regards Bram --------------------------------------------------------------------- The official User-To-User support forum of the Apache HTTP Server Project. See <URL:http://httpd.apache.org/userslist.html> for more info. To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx " from the digest: users-digest-unsubscribe@xxxxxxxxxxxxxxxx For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx
|