Re: [RFC/PATCH] Add sysfs control to modify a user's cpu share

[Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>]

On Tue, Oct 02, 2007 at 06:12:39PM -0400, Eric St-Laurent wrote:
> While a sysfs interface is OK and somewhat orthogonal to the interface
> proposed the containers patches, I think maybe a new syscall should be
> considered.

We had discussed syscall vs filesystem based interface for resource
management [1] and there was a heavy bias favoring filesystem based interface,
based on which the container (now "cgroup") filesystem evolved.

Where we already have one interface defined, I would be against adding 
an equivalent syscall interface.

Note that this "fair-user" scheduling can in theory be accomplished
using the same cgroup based interface, but requires some extra setup in
userspace (either to run a daemon which moves tasks to appropriate
control groups/containers upon their uid change OR to modify initrd to mount 
cgroup filesystem at early bootup time). I expect most distros to enable
CONFIG_FAIR_CGROUP_SCHED (control group based fair group scheduler) and not 
CONFIG_FAIR_USER_SHCED (user id based fair group scheduler). The only
reason why we are providing CONFIG_FAIR_USER_SCHED and the associated
sysfs interface is to help test group scheduler w/o requiring knowledge
of cgroup filesystem.

Reference:

1. http://marc.info/?l=linux-kernel&m=116231242201300&w=2

-- 
Regards,
vatsa