[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] Should we mark RTDS as supported feature from experimental feature?



On Wed, Apr 27, 2016 at 8:27 AM, Dario Faggioli
<dario.faggioli@xxxxxxxxxx> wrote:
> On Tue, 2016-04-26 at 21:16 -0400, Meng Xu wrote:
>> > It's indeed not a must for real-time schedulers. In fact, it's only
>> > important if one wants the system to be overall usable, when using
>> > a
>> > real-time scheduler. :-P
>> >
>> > Also, I may be wrong but it should not be too hard to implement...
>> > I.e., a win-win. :-)
>> I'm thinking if we want to implement work-conserving policy in RTDS,
>> how should we allocate the unused resource to domains. Should this
>> allocation be promotional to the budget/period each domain is
>> configured with?
>> I guess the complexity totally depends on which work-conserving
>> algorithm we want to encode into RTDS.
>>
> Indeed it does.
>
> Everything works for me, basically. As you say, it would not be a
> critical aspect of this scheduler, and the implementation details of
> the work conserving mode is not going to be the reason why people
> choose it anyway... It's just to avoid that people runs away from it
> (and from Xen) screaming! :-)

I see. Right! This is a good point.

>
> So, for instance, how do you manage non real-time VMs in RT Xen?

RT-Xen is not working-serving right now. The way we handle the non RT
VM in RT-Xen 2.1 (not the latest version) is that we use another bit
in rt_vcpu to indicate if a VCPU is RT or not.

The non-RT VCPUs always have lower priority than the RT VCPUs.

> You
> say you still use EDF, how do you do that?

When RT VCPUs all depleted budget,  the non-RT VCPUs will be scheduled
by gEDF scheduling policy.

> When does one non real-time
> VM preempt another non real-time VM? (Ideally, I'd go and check the RT-
> Xen code that does this myself, but right now, I can't, sorry.)

The non-RT VCPU cannot be scheduled if any RT VCPU still has budget.
Once non-RT VCPUs are scheduled, they are preempted/scheduled based on
gEDF, since a non-RT VCPU also has budget and period parameters.

>
> We could go for this that you have already, and as soon as a VM
> exhausts its budget, we demote it to non real-time, until it receives
> the replenishment. Or something like that.

Right. To make it work-conserving, we will have to keep decreasing the
priority whenever it runs out of budget at that priority, until there
is no idle resource in the system any more.

>
> In this case, we basically get two features at the cost of one (support
> for non real-time VMs and work conserving mode for real-time VMs). Not
> to mention that you basically have the code already, and "only" need to
> upstream it! :-DD
>

Right. That is true... Let me think about it and send out a design later.

Meng


-- 
-----------
Meng Xu
PhD Student in Computer and Information Science
University of Pennsylvania
http://www.cis.upenn.edu/~mengxu/

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.