Re: [ckrm-tech] [PATCH 0/4] sched: Add CPU rate caps

[Balbir Singh]

Peter Williams wrote:
> Balbir Singh wrote:
>
> > Peter Williams wrote:
> >
> > > Balbir Singh wrote:
> > >
> > > > Peter Williams wrote:
> >
> > <snip>
> >
> > > > Is it possible that the effective tasks
> > > > is greater than the limit of the group?
> > >
> > >
> > >
> > > Yes.
> > >
> > > > How do we handle this scenario?
> > >
> > >
> > >
> > > You've got the problem back to front.  If the number of effective 
> > > tasks is less than the group limit then you have the situation that 
> > > needs special handling (not the other way around).  I.e. if the 
> > > number of effective tasks is less than the group limit then (strictly 
> > > speaking) there's no need to do any capping at all as the demand is 
> > > less than the limit.  However, in the case where the group limit is 
> > > less than one CPU (i.e. less than 1000) the recommended thing to do 
> > > would be set the limit of each task in the group to the group limit.
> > >
> > > Obviously, group limits can be greater than one CPU (i.e. 1000).
> > >
> > > The number of CPUs on the system also needs to be taken into account 
> > > for group capping as if the group cap is greater than the number of 
> > > CPUs there's no way it can be exceeded and tasks in this group would 
> > > not need any processing.
> >
> > What if we have a group limit of 100 (out of 1000) and 150 effective 
> > tasks in
> > the group? How do you calculate the cap of each task?
>
>
> Personally I'd round up to 1 :-) but rounding down to zero is also an 
> option.  The reason I'd opt for 1 is that a zero cap has a special 
> meaning i.e. background.
>
> > I hope my understanding of effective tasks is correct.
>
>
> Yes, but I think that you fail to realize that this problem (a lower 
> limit to what caps can be enforced) exists for any mechanism due to the 
> fact we're stuck with discrete mathematics in computers.  This includes 
> floating point representations of numbers which are just crude (discrete 
> maths) approximations of real numbers.

I do appreciate and realize the problem, thats why I asked the question.

There are some ways of solving this problem (that I could think about)

1. Keep a whole number and fraction pair and increment the fraction until
  it reaches a whole number and then schedule the task when the whole
  number value reaches a minimal threshold. Or provide tasks with some
  minimal whole number ticks in advance and then do not schedule them
  again till their fractions add up to the whole number (credit system).

  For example if T1 and T2 have a cap of 0.5%. Then represent the values
  as whole number zero and fraction represented as 1 and divisor as 2.

  Every two ticks their whole number would become 1 and fraction 0, divisor 2.
  Schedule the tasks for a tick whenever its whole number becomes 1
  and reset then its whole number to 0.


2. In a group based cap management system, schedule some tasks (highest priority)
  until their cap run out. In the subsequent rounds pick and choose tasks that
  did not get a chance to run earlier.

Solving this is indeed a interesting problem.

> Peter


--

	Balbir Singh,
	Linux Technology Center,
	IBM Software Labs
-
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/