DragonFly BSD
DragonFly kernel List (threaded) for 2003-09
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: Anybody working on removing sendmail from base?


From: sander <sander@xxxxxxxxxxxxxxxxxxx>
Date: Sun, 28 Sep 2003 12:56:50 +0300 (EEST)

On Sun, 28 Sep 2003, Matthew Dillon wrote:

>     I really do think that we will be able to use some sort of VFS layering
>     feature to deal with default compiler selection and things like that.
>     Don't think of it so much as actually creating a VFS, instead think of it
>     as an abstraction many forms of which can be optimized down to a few
>     permanent entries in the namecache.  The /etc/rc* system would set up
>     the entries when the system boots and they could also be changed at run
>     time.
>

So this would essentialy be a process subtree specific union mount of
/usr/bin and nano-filesystem conataining just alternate entries for gcc,
cc, etc?

>     Then consider the ability to 'stack' the abstraction ('stack' virtualized
>     VFS layers) to create environments.  e.g. user A might want GCC3.2 as
>     the default while user B might want GCC2.95.4.
>

sure - if you can do process subtree specific stacking / mounting this
is simple.

>     Now, if you can picture all of that, go back to the idea of a VFS
>     'process' running in userland, but then consider that all such a process
>     really has to do is manage entries in the namecache and would only be
>     messaged when the kernel can't find the request in the cache.  We still
>     wind up with almost 100% the performance that we would have with the
>     underlying native filesystem.
>

Wouldn't it be more efficent for a selection of such cases if you just
adopted variant symlinks? You would then have

	/usr/bin/gcc -> /usr/compilers/${gcc_ver}/bin/gcc

and where it points to would depend directly on environment.

>     Anything we can do within the kernel's namecache is basically going to be
>     freebee from a performance standpoint.
>

you have to differentiate between N versions that are in various ways
associated with calling processes though.

> 						-Matt
>


	Sander

+++ Out of cheese error +++





[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]