VM and jailed processes

[Andrew]

On Sun, 2006-05-14 at 10:01 -0400, Bill Moran wrote:
> Andrew <andrew.chace at gmail.com> wrote:
> 
> > Ok, I'm a bit fuzzy on some of the details, so take it easy. ;-)
> > 
> > It's my understanding that if there is more than one instance of a
> > specific application running, then portions of the code are shared in
> > memory. I would assume that would apply to dynamically linked
> > applications as well; i.e. if two different applications are linked
> > against the same library, the given code exists in only one location in
> > memory. Is this correct?
> > 
> > The second portion of my question is, how does this apply to jailed
> > processes? Looking through the architecture handbook, I did not see any
> > references to VM, which leads me to believe that the standard rules
> > apply to jails as well. So, for instance, if I was to provide a hosting
> > service with numerous instances of Apache running in individual jails,
> > could I assume that base memory usage (ie idle, not serving requests)
> > would increase at a roughly linear rate?
> 
> Keep in mind that if you set up jails the cononical way, each jail instance
> will have it's own installation of Apache.  Even if each of these
> installations are _identical_, they're still seperate, and the kernel
> has now way of knowing that /jail1/usr/bin/httpd and /jail2/usr/bin/httpd
> are the same execution image (Unless you're doing symlinks or hardlinks).
> 
> So getting that kind of memory sharing will require some extra work on
> your part, above and beyond what is normally done for a jail.
> 

Hi Bill,

I'm thinking of using mount_nullfs(8) to provide read-only mounts for
all the executables in each jail. I've been doing some reading, 'man
rtld(1)', and it seems that the linker will take of sharing non-writable
code between processes, even if the executables are loaded from
different mount-points/file-systems. 

But thanks for the heads up...

-Andrew