From: "Serge E. Hallyn" <serue@us.ibm.com>
To: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Paul Menage <menage@google.com>,
ebiederm@xmission.com, sam@vilain.net, akpm@linux-foundation.org,
pj@sgi.com, dev@sw.ru, xemul@sw.ru, serue@us.ibm.com,
containers@lists.osdl.org, winget@google.com,
ckrm-tech@lists.sourceforge.net, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 0/2] resource control file system - aka containers on top of nsproxy!
Date: Wed, 7 Mar 2007 11:43:46 -0600 [thread overview]
Message-ID: <20070307174346.GA19521@sergelap.austin.ibm.com> (raw)
In-Reply-To: <20070307173031.GC2336@in.ibm.com>
Quoting Srivatsa Vaddagiri (vatsa@in.ibm.com):
> On Tue, Mar 06, 2007 at 06:32:07PM -0800, Paul Menage wrote:
> > I'm not really sure that I see the value of having this be part of
> > nsproxy rather than the previous independent container (and
> > container_group) structure.
>
> *shrug*
>
> I wrote the patch mainly to see whether the stuff container folks (Sam Vilain
> et al) were complaining abt (that container structure abstraction
> inside the kernel is redundant/unnecessary) made sense or not.
I still think the complaint was about terminology, not implementation.
They just didn't want you calling them containers.
> The rcfs patches demonstrate that it is possible to implement resource control
> on top of just nsproxy -and- give the same interface that you now
> have. In essense, I would say that the rcfs patches are about 70% same as your
> original V7 container patches.
>
> However as I am converting over cpusets to work on top of nsproxy, I
> have learnt few things:
>
> container structure in your patches provides for these things:
>
> a. A way to group tasks
> b. A way to maintain several hierarchies of such groups
>
> If you consider just a. then I agree that container abstraction is
> redundant, esp for vserver resource control (nsproxy can already be used
> to group tasks).
>
> What nsproxy doesn't provide is b - a way to represent hierarchies of
> groups.
>
> So we got several choices here.
>
> 1. Introduce the container abstraction as is in your patches
> 2. Extend nsproxy somehow to represent hierarchies
> 3. Let individual resource controllers that -actually- support
> hierarchical resource management maintain hierarchy in their code.
>
> In the last option, nsproxy still is unaware of any hierarchy. Some of
> the resource objects it points to (for ex: cpuset) may maintain a
> hierarchy. For ex: nsproxy->ctlr_data[cpuset_subsys.subsys_id] points to
> a 'struct cpuset' structure which could maintains the hierarchical
> relationship among cpuset objects.
>
> If we consider that most resource controllers may not implement hierarchical
> resource management, then 3 may not be a bad compromise. OTOH if we
> expect *most* resource controllers to support hierarchical resource
> management, then we could be better of with option 1.
>
> Anyway, summarizing on "why nsproxy", the main point (I think) is about
> using existing abstraction in the kernel.
But nsproxy is not an abstraction, it's an implementation
detail/optimization. I'm mostly being quiet because i don't
particularly care if it gets expanded upon, but it's nothing more than
that right now.
> > As far as I can see, you're putting the
> > container subsystem state pointers and the various task namespace
> > pointers into the same structure (nsproxy) but then they're remaining
> > pretty much independent in terms of code.
> >
> > The impression that I'm getting (correct me if I'm wrong) is:
> >
> > - when you do a mkdir within an rcfs directory, the nsproxy associated
> > with the parent is duplicated, and then each rcfs subsystem gets to
> > set a subsystem-state pointer in that nsproxy
>
> yes.
>
> > - when you move a task into an rcfs container, you create a new
> > nsproxy consisting of the task's old namespaces and its new subsystem
> > pointers. Then you look through the current list of nsproxy objects to
> > see if you find one that matches. If you do, you reuse it, else you
> > create a new nsproxy and link it into the list
>
> yes
>
> > - when you do sys_unshare() or a clone that creates new namespaces,
> > then the task (or its child) will get a new nsproxy that has the rcfs
> > subsystem state associated with the old nsproxy, and one or more
> > namespace pointers cloned to point to new namespaces. So this means
> > that the nsproxy for the task is no longer the nsproxy associated with
> > any directory in rcfs. (So the task will disappear from any "tasks"
> > file in rcfs?)
>
> it "should" disappear yes, although I haven't carefully studied the
> unshare requirements yet.
>
> > You seem to have lost some features, including fork/exit subsystem callbacks
>
> That was mainly to keep it simple for a proof-of-concept patch! We can add it
> back later.
>
> > >What follows is the core (big) patch and the cpu_acct subsystem to serve
> > >as an example of how to use it. I suspect we can make cpusets also work
> > >on top of this very easily.
> >
> > I'd like to see that. I suspect it will be a bit more fiddly than the
> > simple cpu_acct subsystem.
>
> I am almost done with the conversion. And yes cpuset is a beast to
> convert over! Will test and send the patches out tomorrow.
>
> --
> Regards,
> vatsa
next prev parent reply other threads:[~2007-03-07 17:45 UTC|newest]
Thread overview: 125+ messages / expand[flat|nested] mbox.gz Atom feed top
2007-03-01 13:35 [PATCH 0/2] resource control file system - aka containers on top of nsproxy! Srivatsa Vaddagiri
2007-03-01 13:45 ` [PATCH 1/2] rcfs core patch Srivatsa Vaddagiri
2007-03-01 16:31 ` Serge E. Hallyn
2007-03-01 16:46 ` Srivatsa Vaddagiri
2007-03-02 5:06 ` [ckrm-tech] " Balbir Singh
2007-03-03 9:38 ` Srivatsa Vaddagiri
2007-03-08 3:12 ` Eric W. Biederman
2007-03-08 9:10 ` Paul Menage
2007-03-09 0:38 ` Herbert Poetzl
2007-03-09 9:07 ` Kirill Korotaev
2007-03-09 13:29 ` Herbert Poetzl
2007-03-09 17:57 ` Srivatsa Vaddagiri
2007-03-10 1:19 ` Herbert Poetzl
2007-03-11 16:36 ` Serge E. Hallyn
2007-03-12 23:16 ` Herbert Poetzl
2007-03-08 10:13 ` Srivatsa Vaddagiri
2007-03-09 0:48 ` Herbert Poetzl
2007-03-09 2:35 ` Paul Jackson
2007-03-09 9:23 ` Kirill Korotaev
2007-03-09 9:38 ` Paul Jackson
2007-03-09 13:21 ` Herbert Poetzl
2007-03-11 17:09 ` Kirill Korotaev
2007-03-12 23:00 ` Herbert Poetzl
2007-03-13 8:28 ` Kirill Korotaev
2007-03-13 13:55 ` Herbert Poetzl
2007-03-13 14:11 ` [ckrm-tech] " Srivatsa Vaddagiri
2007-03-13 15:52 ` Herbert Poetzl
2007-03-09 18:14 ` Srivatsa Vaddagiri
2007-03-09 19:25 ` Paul Jackson
2007-03-10 1:00 ` Herbert Poetzl
2007-03-10 1:31 ` Paul Jackson
2007-03-10 0:56 ` Herbert Poetzl
2007-03-09 16:16 ` Serge E. Hallyn
2007-03-01 13:50 ` [PATCH 2/2] cpu_accounting controller Srivatsa Vaddagiri
2007-03-01 19:39 ` [PATCH 0/2] resource control file system - aka containers on top of nsproxy! Paul Jackson
2007-03-02 15:45 ` Kirill Korotaev
2007-03-02 16:52 ` Andrew Morton
2007-03-02 17:25 ` Kirill Korotaev
2007-03-03 17:45 ` Herbert Poetzl
2007-03-03 21:22 ` Paul Jackson
2007-03-05 17:47 ` Srivatsa Vaddagiri
2007-03-03 9:36 ` Srivatsa Vaddagiri
2007-03-03 10:21 ` Paul Jackson
2007-03-05 17:02 ` Srivatsa Vaddagiri
2007-03-03 17:32 ` Herbert Poetzl
2007-03-05 17:34 ` [ckrm-tech] " Srivatsa Vaddagiri
2007-03-05 18:39 ` Herbert Poetzl
2007-03-06 10:39 ` Srivatsa Vaddagiri
2007-03-06 13:28 ` Herbert Poetzl
2007-03-06 16:21 ` Srivatsa Vaddagiri
2007-03-07 2:32 ` Paul Menage
2007-03-07 17:30 ` Srivatsa Vaddagiri
2007-03-07 17:29 ` Paul Menage
2007-03-07 17:52 ` [ckrm-tech] " Srivatsa Vaddagiri
2007-03-07 17:32 ` Srivatsa Vaddagiri
2007-03-07 17:43 ` Serge E. Hallyn [this message]
2007-03-07 17:46 ` Paul Menage
2007-03-07 23:16 ` Eric W. Biederman
2007-03-08 11:39 ` Srivatsa Vaddagiri
2007-03-07 18:00 ` Srivatsa Vaddagiri
2007-03-07 20:58 ` Serge E. Hallyn
2007-03-07 21:20 ` Paul Menage
2007-03-07 21:59 ` Serge E. Hallyn
2007-03-07 22:13 ` Dave Hansen
2007-03-07 23:13 ` Eric W. Biederman
2007-03-12 14:11 ` [ckrm-tech] " Srivatsa Vaddagiri
2007-03-07 22:32 ` Eric W. Biederman
2007-03-07 23:18 ` Paul Menage
2007-03-08 0:35 ` Sam Vilain
2007-03-08 0:42 ` Paul Menage
2007-03-08 0:53 ` Sam Vilain
2007-03-08 0:58 ` [ckrm-tech] " Paul Menage
2007-03-08 1:32 ` Eric W. Biederman
2007-03-08 1:35 ` Paul Menage
2007-03-08 2:25 ` Eric W. Biederman
2007-03-09 0:56 ` Herbert Poetzl
2007-03-09 0:53 ` Herbert Poetzl
2007-03-09 18:19 ` Srivatsa Vaddagiri
2007-03-09 19:36 ` Paul Jackson
2007-03-09 21:52 ` Herbert Poetzl
2007-03-09 22:06 ` Paul Jackson
2007-03-12 14:01 ` Srivatsa Vaddagiri
2007-03-12 15:15 ` Srivatsa Vaddagiri
2007-03-12 20:26 ` Paul Jackson
2007-03-09 4:30 ` Paul Jackson
2007-03-08 2:47 ` Sam Vilain
2007-03-08 2:57 ` Paul Menage
2007-03-08 3:32 ` Sam Vilain
2007-03-08 6:10 ` Matt Helsley
2007-03-08 6:44 ` Eric W. Biederman
2007-03-09 1:06 ` Herbert Poetzl
2007-03-10 9:06 ` Sam Vilain
2007-03-11 21:15 ` Paul Jackson
2007-03-12 9:35 ` Sam Vilain
2007-03-12 10:00 ` Paul Menage
2007-03-12 23:21 ` Herbert Poetzl
2007-03-13 2:25 ` Paul Menage
2007-03-13 15:57 ` Herbert Poetzl
2007-03-09 4:37 ` Paul Jackson
2007-03-08 6:32 ` Eric W. Biederman
2007-03-08 9:10 ` Paul Menage
2007-03-09 16:50 ` Serge E. Hallyn
2007-03-22 14:08 ` Srivatsa Vaddagiri
2007-03-22 14:39 ` Serge E. Hallyn
2007-03-22 14:56 ` Srivatsa Vaddagiri
2007-03-09 4:27 ` Paul Jackson
2007-03-10 8:52 ` Sam Vilain
2007-03-10 9:11 ` Paul Jackson
2007-03-09 16:34 ` Srivatsa Vaddagiri
2007-03-09 16:41 ` [ckrm-tech] " Srivatsa Vaddagiri
2007-03-09 22:09 ` Paul Menage
2007-03-10 2:02 ` Srivatsa Vaddagiri
2007-03-10 3:19 ` [ckrm-tech] " Srivatsa Vaddagiri
2007-03-12 15:07 ` Srivatsa Vaddagiri
2007-03-12 15:56 ` Serge E. Hallyn
2007-03-12 16:20 ` Srivatsa Vaddagiri
2007-03-12 17:25 ` Serge E. Hallyn
2007-03-12 21:15 ` Sam Vilain
2007-03-12 23:31 ` Herbert Poetzl
2007-03-13 2:22 ` Srivatsa Vaddagiri
2007-03-08 0:50 ` Sam Vilain
2007-03-08 11:30 ` Srivatsa Vaddagiri
2007-03-09 1:16 ` Herbert Poetzl
2007-03-09 18:41 ` Srivatsa Vaddagiri
2007-03-10 2:03 ` Herbert Poetzl
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20070307174346.GA19521@sergelap.austin.ibm.com \
--to=serue@us.ibm.com \
--cc=akpm@linux-foundation.org \
--cc=ckrm-tech@lists.sourceforge.net \
--cc=containers@lists.osdl.org \
--cc=dev@sw.ru \
--cc=ebiederm@xmission.com \
--cc=linux-kernel@vger.kernel.org \
--cc=menage@google.com \
--cc=pj@sgi.com \
--cc=sam@vilain.net \
--cc=vatsa@in.ibm.com \
--cc=winget@google.com \
--cc=xemul@sw.ru \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).