Re: [Xen-devel] [RFC 2/6] schedule: account true system idle time

[Dario Faggioli <dfaggioli@xxxxxxxx>]

On Fri, 2019-07-26 at 15:42 +0300, Andrii Anisov wrote:
> Hello Dario,
> 
Hi,

> On 26.07.19 15:00, Dario Faggioli wrote:
> > On Fri, 2019-07-26 at 13:37 +0300, Andrii Anisov wrote:
> > I see how it is rather easy to do something like this, so
> > I understand it being done like this in an RFC patch, but I don't
> > think
> > it's correct.
> 
> This is the VERY RFC with the minimal changes to the existing code
> and adopting existing approaches.
>
Sure, I know, and this is fine.

> This topic is really complex and requires wide discussion, so this
> series is rather an invitation to the discussion.
> 
Absolutely.

> > And, on an even more general perspective, the fact that the
> > hypervisor,
> > when scheduling the idle vcpu, runs softirq, tasklets, etc, it's a
> > generic concept, not an arch specific one. So, we really should
> > find a
> > way to implement this in common code, not in arch code.
> 
> Yes, in terms of this patch, idle_vcpu_runstate_change() better be
> moved to common/schedule.c.
> 
I think we should, first of all, think, if using runstates and
runstates manipulation functions is really the best way forward here.

And, if that reveals to be the case, I feel like runstates would need
to be extended to be able to deal with want we want to achieve.

I'll think more about this, and try to form an idea...

> > Maybe, but I'm just thinking out loud, and I need to think more
> > about
> > this, we can do things the other way round. I.e., we measure the
> > time
> > that it takes to run softirq and tasklets, and we subtract it from
> > idle_vcpu runtime?
> 
> In the patch "schedule: account all the hypervisor time to the idle
> vcpu" I extend what I think should be accounted for the hypervisor
> run time. And subtraction approach will result in more complex code
> over there.
> 
Yep, I quickly looked at it, but I need to review it carefully, to
properly understand it, come up with comments, think about
alternatives, etc.

Will do that soon, hopefully.

For now, just consider that, IMO, the big value of this series (or, if
you want, the discussion this series is aimed at starting) is getting
Xen toward having a better, more accurate and more fine grained time
accounting and reporting.

If we time. e.g., interrupts, softirqs and tasklets, we can store such
metrics too, and, if we want, report a breakout of the time spent in
hypervisor... something like (in xentop, as you're doing already, or
somewhere else):

 hyp=12%(irq=4%+softirq=1%+tasklet=5%+other=2%) 

Anyway, thanks again and Regards
-- 
Dario Faggioli, Ph.D
http://about.me/dario.faggioli
Virtualization Software Engineer
SUSE Labs, SUSE https://www.suse.com/
-------------------------------------------------------------------
<<This happens because _I_ choose it to happen!>> (Raistlin Majere)

Attachment: signature.asc
Description: This is a digitally signed message part

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel