Re: [patch 1/3] add basic accounting fields to taskstats

[Balbir Singh <balbir@in.ibm.com>]

Jay Lan wrote:
> Jay Lan wrote:
>
> [snip]
>
> > > > +    /* Each process gets a minimum of a half tick cpu time */
> > > > +    if ((stats->ac_utime == 0) && (stats->ac_stime == 0)) {
> > > > +        stats->ac_stime = USEC_PER_TICK/2;
> > > > +    }
> > > > +
> > >
> > >
> > >
> > > This is confusing. Half tick does not make any sense from the
> > > scheduler view point (or am I missing something?), so why
> > > return half a tick to the user.
> >
> >
> > It must be inherited from old code dated back to Cray UNICOS.
> > I do not know if bad thing can happen if both utime and stime
> > are less than 1 usec...  I guess not. But i agree that
> > half a tick does not make sense. To play safe, we can change
> > it to 1 usec if both utime and stime are sub microsecond.
> > What do you think?
>
> Hi Balbir,
>
> I figured this out. The tsk->stime (and utime as well) are
> charged by 1 tick (or cputime) from the timer interrupt handler
> through update_process_times->account_{user,system}_time.
>
> The clock resolution is a tick. Any short process less than
> 1 tick will the counter being 0. It can be from 0 to 0.99999...
> tick. A half tick is the average value.

But the scheduling happens in the granularity of a tick, so the minimum each 
task gets is a tick.

> I think it makes more sense to assign a half tick than assign
> 1 usec to the stime. What do you think? Certainly the code need
> better explanation.

Can't we leave these values as zero in case both stime and utime are zero.


> Regards,
>  - jay
>
>
> [snip]


--
	Warm Regards,
	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@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/