Re: [PATCH 13/15] sched: add bandwidth management for sched_dl.

[Juri Lelli]

On 05/29/2012 11:58 AM, Peter Zijlstra wrote:
> On Sat, 2012-05-26 at 13:07 +0200, Juri Lelli wrote:
> > Hi,
> >
> > On 05/25/2012 12:38 PM, Peter Zijlstra wrote:
> > > On Wed, 2012-05-23 at 23:42 +0200, Juri Lelli wrote:
> > > > +/*
> > > > + * Coupling of -dl and -rt bandwidth.
> > > > + *
> > > > + * Here we check, while setting the system wide bandwidth available
> > > > + * for -dl tasks and groups, if the new values are consistent with
> > > > + * the system settings for the bandwidth available to -rt entities.
> > > > + *
> > > > + * IOW, we want to enforce that
> > > > + *
> > > > + *   rt_bandwidth + dl_bandwidth<= 100%
> > > > + *
> > > > + * is always true.
> > > > + */
> > >
> > > I was thinking we could do it the other way around, have have
> > > dl_bandwidth included in rt_bandwidth.
> >
> > If I understand correctly, you are proposing to treat -dl tasks as a
> > special case of "real-time" tasks. Then we could reserve some bw to
> > "real-time" (rt_bandwidth cap) activities and give a piece of this
> > bw to -dl tasks (what remains is for -rt tasks). This is in principle
> > nice and useful, but I'm not quite sure that this is the right point
> > to achieve this logical behavior.
> > I mean, -dl and -rt tasks are separately treated, so it is probably
> > cleaner to manage their knobs separately. They have to coexist rather
> > than be considered one a sub-case of the other. A better way to go
> > for a common cap for them is probably the (long-term) hierarchical
> > scheduling mechanism.
> >
> > So, I would prefer to keep the interface as is for now, but I can also
> > completely misunderstood your thoughts :-P.
>
> The thing is, keeping it separate makes for an impossible configuration
> scenario. Esp. once we enable !root usage. The proposed 5% is very
> limiting and regular users won't have sufficient privilege to change it.

Ok, now I understand your point better, and I agree that 5% is hardly
usable for !root users. However, I also think this is probably more a
system admin problem. I mean, a sys admin that wants his users to play
with -deadline scheduling should have thought how to properly set up his
system, and the fact that something must be configured by hand to give
users a usable system is generally not a so bad idea.

> Also lowering FIFO/RR by default isn't a real option since people expect
> that to get all time already (however silly that expectation is).

I agree. Don't want to spoil users expectations :-).

Thanks and regards,

- Juri
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/