Naming the "Task containers" framework

Paul Menage menage at google.com
Tue Sep 11 07:38:57 PDT 2007


On 9/11/07, Serge E. Hallyn <serue at us.ibm.com> wrote:
> Quoting Paul Menage (menage at google.com):
> > At the mini-summit, and at other times, I've heard the repeated
> > complaint that having the word "container" in the name of the "task
> > container" framework leads to ambiguity. And separately from the
> > complaints, I've seen the awkwardness that people end up with when
> > they feel they have to distinguish between the "containers" abstract
> > concept, and "Paul's containers" ...
> >
> > With the hope/prospect of having the framework merged some time after
> > the kernel summit, I guess now's a good time to bow to the pressure
> > and find some compromise that everyone likes, before we actually hit
> > mainline. (Maybe earlier would have been even better, but ...)
> >
> > Of the various possible names that have been suggested, there are a
> > couple that (to me) stand out as good options:
> >
> > - control groups
> > - task sets
> >
> > The former (coined by Eric during a brainstorming session yesterday)
> > seems to capture the enforcement aspect of the framework (sysadmin can
> > use it to control the behaviour of processes, processes can't escape
> > from groups), without suggesting that it can only be used for resource
> > controllers (as some alternative names such as "resource groups"
> > imply) and would be a choice that I could be happy with.
> >
> > Does anyone have strong views on other alternative names (or even the
> > idea of keeping "task containers")?
>
> Purely subjectively I prefer control groups, but task sets is more
> descriptive about the implementation.

I described it as "control groups" during the kernel summit
presentation and no-one seemed to object to that name.

As far as I can tell, the general mood seems to be in favour of
"control groups" - no-one else has expressed a preference for "task
sets".

I think it's more important to express the overall intention of the
feature rather than the implementation. Ted T'so (I think) asked at
the summit whether things other than tasks could be first-class
members of control groups - I said that currently they can't, but
people may find interesting ways to conveniently make non-task objects
first-class members in the future (rather than just holding reference
counts the way pages do currently).

>
> So I'd have to vote for task sets.
>
> I like 'task containers', but it really is a pain trying to keep clear
> which containers I'm talking about from one sentence to the next.

Right, enough people have said this to me now, and I've seen the
awkwardness that it entails - I don't want to add a subsystem to the
kernel that's forever referred to as "Paul's containers" :-)

Paul


More information about the Containers mailing list