From: Chandra Seetharaman (sekharan_at_us.ibm.com)
Date: Fri 07 Nov 2003 - 19:46:27 GMT
On Fri, Nov 07, 2003 at 04:44:17PM +0000, Sam Vilain wrote:
Hi Sam,
Below are my comments.
> On Fri, 07 Nov 2003 03:17, Chandra Seetharaman wrote;
>
> My, you've picked the same SysCall number as the one that the Linux
> Vserver project has reserved!
It was just a coincidence... I was woking off test5 and haven't looked at
test9 :(... I 'll change it when we move forward.
>
> What a co-incidence, because we should be working together as we're
> approaching facets of the same problem :-). We do virtualisation, you
> do resource management.
>
> We have 273 set up as a Virtual Server `syscall switch'. Our current
> stable development stream that includes the switch[1] is against
> 2.4.x. Take a look at how we've set up the syscall switch - it's
> extensible and forward thinking, and could easily accommodate CKRM's
> calls.
I looked at your project pages, and in my opinion, both of the projects
have their own value independent of each other, using them together will
give more control to virtual server administrators over how the resources
are used by different security contexts.
But, integrating CKRM and vserver might not add more value and provide
unneccessary burden for somebody that want sees value in only one of them.
>
> We currently use this syscall for virtualisation purposes; things like
> hiding processes, blocking inter-context signals, limiting bind(),
> etc.
>
> We've also got a CPU scheduler that works for the O(1) kernel, that
> allows you to set the desired CPU usage level for a context with soft
> limits, along with a syscall to adjust the priority [2].
>
> Is there any chance of co-operation here? If you will let us extend
> CKRM for virtualisation, we can share the syscall and end up with a
> super-duper virtualisation/resource management system for Linux! And
> SysAdmins will appreciate coherance between the numbers used for each.
With CKRM design this can be easily done. RBCE can be extended to allow
rules that can be defined to classify tasks based on their security
context. And by limiting the resource shares used by each class, one
can control how much resource a particular security context uses.
Comments anybody ?
chandra
>
> 1.
> http://www.13thfloor.at/vserver/d_release/v1.1.0/patch-2.4.23-pre9-vs1.1.0.diff
> 2. http://www.vilain.net/linux/ctx/split-2.4.22-ac4-c17g2/
> --
> Sam Vilain, sam_at_vilain.net
>
> If youve seen one redwood, youve seen them all.
> RONALD REAGAN
>
>
------------------------------------------------------------------------ Chandra Seetharaman | Be careful what you choose.... - sekharan_at_us.ibm.com | .......you may get it. ---------------------------------------------------------------------- _______________________________________________ Vserver mailing list Vserver_at_list.linux-vserver.org http://list.linux-vserver.org/mailman/listinfo/vserver