From: Herbert P÷tzl (herbert_at_13thfloor.at)
Date: Thu 10 Jul 2003 - 03:33:23 BST
On Thu, Jul 10, 2003 at 02:12:31AM +0100, Sam Vilain wrote:
> Hi all,
> Linus has been threatening  recently to begin the 2.6.0-pre series.
> With this in mind, I think that it is critical that the vserver
> community makes an effort to try and get vserver adopted in this
I guess as one of the protagonists of vserver I'm
allowed to disbelieve that this will happen ...
> Everyone can help - even the lazy or benevolant; what I'm thinking is
> that a well presented petition from users sent to the Linux Kernel
> Mailing List will assist in rallying support for the notion.
> Statistics from how active the project web site has been, and how many
> downloads of the software there were in the last X months would also
> be good (Jacques?).
> The other thing, of course, is porting the patches to the current 2.5
> series. It would help if each piece of the patch could be seperated
> logically (Alex, Herbert, I know one of you has already done some work
I've done some categorizing of Alexey's
rh-kernel-2.4.18-la-1054303950/ by splitting
up each hunk into a directory/filename combo
and manually reviewing and categorizing each
of those parts ... I guess it is outdated.
> on this). We should have a serious look at merging Alex's & Jacques'
> trees, along with deciding on the best way to handle some of the
> fundamental changes that Herbert's patches introduced.
In my opinion, the best solution would be to
throw away almost everything we've written so
far, only keeping the concepts and calling the
existing patches a working prototype ...
I know this sounds very fatalistic, but I'll try
to give reasons for that later on ...
> Then, we can look at porting each piece to the 2.5.* tree, and
> including proper documentation and Configure options where relevant.
> I don't know if you've come across this before, but if you supply the
> -D DEFINENAME option to `patch', it will insert C-like #ifdef
> DEFINENAME statements around each change to the source code. This can
> assist in creating kernel Configure options for the code.
and I'm 100% sure that nobody around Linus or
Linus himself will accept a patch with 200++ #ifdefs,
I wouldn't either ...
> I'm happy to co-ordinate the patching effort if no-one else can be
> bothered, but I do not have the time available to do all the merging
> by myself and debug it thoroughly. I know that at least the scheduler
> will require my personal involvement :-). If someone already running
> a vserver page wishes to set up the petition, that would be good too -
> otherwise I could set one up.
>From my point of view, the following issues
still have to be addressed ...
- separation of namespace
- complete virtual network (no alias stuff)
- complete virtual capability set
- virtualized pids/pgs/[tp]tys
- real init task (a bonus of the previous)
- a good resource limit solution
- advanced scheduler (virtual)
- a really good quota solution ;)
and the project has to be broken down into
independant _and_ useful patches, which can
in turn be fed to the Linus tree ...
- virtual task context
- virtual network devices
- immutable link inversion
- resource limits (mem/tasks)
- virtual scheduler
- subtree/virtual quota
do not get me wrong, of course I will help to
strive for this higher goal ...
> Sam Vilain, sam_at_vilain.net
> If youve seen one redwood, youve seen them all.
> RONALD REAGAN