All of lore.kernel.org
 help / color / mirror / Atom feed
* cap and reservation functionality in the credit2 scheduler?
@ 2011-10-05  4:43 Eunbyung Park
  2011-10-05 10:24 ` George Dunlap
  0 siblings, 1 reply; 5+ messages in thread
From: Eunbyung Park @ 2011-10-05  4:43 UTC (permalink / raw)
  To: george.dunlap, xen-devel

As far as I know, from the up-to-date xen-unstable, the credit2
scheduler lacks the functionalities of cap and reservation.

Is there any on-going work for this?

I would really expect the functions to come into us. :)

>From my humble point of view, probably additional timer may be required
to implement them because there isn't any periodic timer in the credit2
unlike the previous credit scheduler.

what do you think about it?

Thanks.

-- 
Best Regards,
Eunbyung Park

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: cap and reservation functionality in the credit2 scheduler?
  2011-10-05  4:43 cap and reservation functionality in the credit2 scheduler? Eunbyung Park
@ 2011-10-05 10:24 ` George Dunlap
  2011-10-05 11:08   ` Eunbyung Park
  0 siblings, 1 reply; 5+ messages in thread
From: George Dunlap @ 2011-10-05 10:24 UTC (permalink / raw)
  To: Eunbyung Park; +Cc: xen-devel

2011/10/5 Eunbyung Park <silverbottlep@gmail.com>:
> As far as I know, from the up-to-date xen-unstable, the credit2
> scheduler lacks the functionalities of cap and reservation.
>
> Is there any on-going work for this?
>
> I would really expect the functions to come into us. :)

Unfortuantely not -- the credit2 work has been stalled for lack of resources.

> >From my humble point of view, probably additional timer may be required
> to implement them because there isn't any periodic timer in the credit2
> unlike the previous credit scheduler.
>
> what do you think about it?

I'm not opposed in principle, but until I've seen the whole algorithm,
I can't really comment. :-)  Do you have an algorithm in mind?

Unfortunately, I never got around to solving that part.  I was focused
on getting a clean, simple, mostly predictable algorithm to handle
weights.  The algorithm itself may therefore need some tweaking to
make cap / reservation possible.

 -George

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: cap and reservation functionality in the credit2 scheduler?
  2011-10-05 10:24 ` George Dunlap
@ 2011-10-05 11:08   ` Eunbyung Park
  2011-10-06 13:57     ` George Dunlap
  0 siblings, 1 reply; 5+ messages in thread
From: Eunbyung Park @ 2011-10-05 11:08 UTC (permalink / raw)
  To: George Dunlap; +Cc: xen-devel

2011-10-05 오후 7:24, George Dunlap 쓴 글:
>> > From my humble point of view, probably additional timer may be required
>> to implement them because there isn't any periodic timer in the credit2
>> unlike the previous credit scheduler.
>>
>> what do you think about it?
> I'm not opposed in principle, but until I've seen the whole algorithm,
> I can't really comment. :-)  Do you have an algorithm in mind?
Actually, I don't have any idea in mind yet.
however, I will try to devise a method to support them and make a patch.
So, If you can afford to review, I will appreciate it :)

And, I think that there are still some to-do lists for the credit2 and 
xl tools needs to include credit2 support.

In addition, I found that adjusting dom0 weight have failed to work, but 
except for the dom0, it worked well.
I found that this problem had been on xen-devel lists long time ago, but 
it seems not to be solved yet.
I think this may be related with spinlock. When I remove the 
vcpu_schedule_lock_irq() in csched_dom_cntl(), it worked well.
I'm not sure what is the real problem because I didn't completely 
understand the mechanism about spinlocks in xen scheduler.

Any comments? or new thread needed?

-- 
Best Regards,
Eunbyung Park

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: cap and reservation functionality in the credit2 scheduler?
  2011-10-05 11:08   ` Eunbyung Park
@ 2011-10-06 13:57     ` George Dunlap
  2011-10-08 12:59       ` Eunbyung Park
  0 siblings, 1 reply; 5+ messages in thread
From: George Dunlap @ 2011-10-06 13:57 UTC (permalink / raw)
  To: Eunbyung Park; +Cc: George Dunlap, xen-devel@lists.xensource.com

On Wed, 2011-10-05 at 12:08 +0100, Eunbyung Park wrote:
> In addition, I found that adjusting dom0 weight have failed to work, 
> but 
> except for the dom0, it worked well.
> I found that this problem had been on xen-devel lists long time ago, but 
> it seems not to be solved yet.
> I think this may be related with spinlock. When I remove the 
> vcpu_schedule_lock_irq() in csched_dom_cntl(), it worked well.
> I'm not sure what is the real problem because I didn't completely 
> understand the mechanism about spinlocks in xen scheduler.

So I'm afraid that instead of telling you how the spinlocks are supposed
to work, I'm going to let you figure out what the locks should be doing
on your own. If you can't do that, then you may not be ready yet to work
on the scheduler.

I hope that doesn't sound too harsh, but the fact is that the scheduler
is probably one of the most deceptively simple parts of the Xen code.
It requires a lot of expertise to know even how to avoid race conditions
and deadlock, not to mention making a simple algorithm which has the
desired effects without undesirable side effects.  (The shadow code is
probably more complicated, for example, but it *looks* more complicated,
so people don't end up in over their head.)  I wouldn't be doing either
you or me a favor by encouraging you to work on something past your
capability level.

> 2011-10-05 오후 7:24, George Dunlap 쓴 글:
> Actually, I don't have any idea in mind yet.
> however, I will try to devise a method to support them and make a patch.
> So, If you can afford to review, I will appreciate it :)

I'll be happy to review whatever you can send to the list. But like I
said above, beware of getting in over your head. :-)

> And, I think that there are still some to-do lists for the credit2 and 
> xl tools needs to include credit2 support.

This should be fairly straightforward to do, and would be much
appreciated!

 -George

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: cap and reservation functionality in the credit2 scheduler?
  2011-10-06 13:57     ` George Dunlap
@ 2011-10-08 12:59       ` Eunbyung Park
  0 siblings, 0 replies; 5+ messages in thread
From: Eunbyung Park @ 2011-10-08 12:59 UTC (permalink / raw)
  To: George Dunlap; +Cc: xen-devel@lists.xensource.com


> I hope that doesn't sound too harsh, but the fact is that the scheduler
> is probably one of the most deceptively simple parts of the Xen code.
> It requires a lot of expertise to know even how to avoid race conditions
> and deadlock, not to mention making a simple algorithm which has the
> desired effects without undesirable side effects.  (The shadow code is
> probably more complicated, for example, but it *looks* more complicated,
> so people don't end up in over their head.)  I wouldn't be doing either
> you or me a favor by encouraging you to work on something past your
> capability level.
It was too harsh comments. just kidding~~:)

Anyway, nowadays, I am working on the project which is related with the 
xen scheduler. Actually, I do not have enough expertise yet, however, I 
am very interested and excited at the xen source code and I am a little 
bit more getting used to it. I hope that I would completely understand 
the all of the mechanisms soon and hack the xen source.

Thanks, george.

--
Best Regards,
Eunbyung Park

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2011-10-08 12:59 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-10-05  4:43 cap and reservation functionality in the credit2 scheduler? Eunbyung Park
2011-10-05 10:24 ` George Dunlap
2011-10-05 11:08   ` Eunbyung Park
2011-10-06 13:57     ` George Dunlap
2011-10-08 12:59       ` Eunbyung Park

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.