Re: [patch] scheduler: rebalance_tick interval update

From: Darren Hart
Date: Mon Nov 15 2004 - 22:52:09 EST


On Tue, 2004-11-16 at 13:27 +1100, Nick Piggin wrote:
>
> Nick Piggin wrote:
>
> >
> > Another example, in some ticks, a CPU won't see the updated 'jiffies',
> > other
> > times it will (at least on Altix systems, this can happen).
> >
> >
>
> Note that if you didn't want to have this rash of balancing attempted after
> a CPU wasn't able to run the rebalance for a long time, the solution would
> be to keep adding the balance interval until it becomes greater than the
> current jiffies.

As I mentioned in my last post, I don't think the "synchronized
rebalancing" is a real concern since the interval isn't likely to be the
same and the CPU_OFFSET macro is already in place to prevent this "rash
of balancing" (nice term :-).

> I actually prefer it to try to make up the lost balances, just from the
> perspective of gathering scheduler statistics.

IMO, scheduler statistics are not worth running load_balance() for no
reason. (And running it two or three times in a row is clearly not
accomplishing anything)

> I don't suspect it happens
> enough to justify adding the extra logic - Darren, are you actually seeing
> problems?

Not seeing in obvious problems, but the existing logic seems incorrect
to me (and the term last_balanced is currently misleading). Running
load_balance() multiple times in order to catch up seems wasteful to me
as well. The current code says something like: run load_balance() 10
times in a second. If the second is almost up and you have only run it
6 times, it will run it 4 times in a row, that just seems wrong to me.


--
Darren Hart <darren@xxxxxxxxxx>

-
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/