VM and jailed processes

[Philip Hallstrom]

>>> 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.

You should also look at ezjail...  it uses the same tricks to reduce the 
size of individual jail systems.  I haven't used it, but keep meaning too 
(next server :)

http://erdgeist.org/arts/software/ezjail/