Subject: Re: [vserver] official kernel inclusion
From: Herbert Poetzl <herbert@13thfloor.at>
Date: Thu, 10 Jan 2008 01:30:25 +0100

On Wed, Jan 09, 2008 at 04:15:16PM -0800, Martin Fick wrote:
> --- Herbert Poetzl <herbert@13thfloor.at> wrote:
> >
> > mainline goes for the heavier process virtualization
> > instead of isolation ...
> 
> Could you explain what you mean by that?  

mainline in general heads for the straight and
complete virtualization approach, they do not bother
with performant isolation and corner cases (yet)

> Are you talking about the recent "PID Namespaces" work
> in the mainline? (bottom of this page:)
> http://lwn.net/Articles/258703/

yep, precisely

> Is this something that is much heavier (I realize it
> is potentially more featureful) than the vserver
> technique?  

nah, not _that_ much heavier, but, in certain cases
too heavy compared to what Linux-VServer is doing.

for example, consider 1000 guests with a single
daemon running inside, on mainline that would give
2000 processes (init + daemon) while we can do with
only 1001 processes (single shared init)

> Now that the mainline has this support,
> would it not make sense, 

> is it possible to use it even for the vserver project?

yes, and we are providing this in _addition_ to
the process isolation, as alternative ... so the
choice (which one you prefer) is yours

best,
Herbert

> -Martin
> 
> 
>       ____________________________________________________________________________________
> Be a better friend, newshound, and 
> know-it-all with Yahoo! Mobile.  Try it now.  http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ