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

Re: [Xen-devel] [PATCH v4 13/15] xl: enable getting and setting soft



On mar, 2013-12-03 at 09:32 +0000, Ian Campbell wrote:
> On Mon, 2013-12-02 at 19:10 +0100, Dario Faggioli wrote:
> > > Is there no option to set both?
> > > 
> > No, not for now. Do you think there should be one?
> 
> Is it never necessary to set the "atomically"? Or is it always possible
> to set one arbitrarily before the other?
> 
From the user perspective (as we're discussing `xl' interface), there
certainly are cases when one want both to have the same or different
values, but it is possible to achieve that by setting them independently
(just takes more time, since you need two `xl vcpu-pin' commands).

Setting soft affinity *never* fails, while setting hard may fail, in the
same exact circumstances when it is failing right now (e.g, setting hard
affinity completely outside of your cpupool). Moreover, setting
whichever one does not interfere with the other (apart from the
possibility of getting a warning, it the system goes through some funny
transient state).

Therefore, I don't think there is an hard requirement for this, although
I'm certainly up for trying to make this more comfortable to use, once
in 4.5.

> > At that point, the most natural
> > thing to do seemed to add a switch to have this deal with soft affinity,
> > which is what happens in this patch.
> > 
> > I of course can come up with pretty much all sort of behavior, e.g.,
> > setting both hard and soft affinity to the same cpumap, if no switch is
> > provided, and then introduce both a '-h' and a '-s', for specifying
> > different cpumaps, at the same time, etc.
> 
> Is setting them both the same pretty much identical in practice to
> setting the hard and leaving the soft at all?
>
Yes it is. That's why I'd rather provide a mechanism for setting both,
but each one to its own value (which could be hard=3-5-soft=all). I
thinks it's more useful.

> I was actually thinking of -b/--both but as you suggest could work too,
> except -h is help so would need a different letter.
> 
Well, sure -h was pretty stupid, sorry. :-)

> Another option would be to take two affinities on the command line, the
> first being hard and the second soft with some special syntax ("-"?) for
> "leave it alone".
> 
Mmm... I think I like this better. So, if we take the approach
implemented in this patch right now, and introduce something like that
later, here's what we'd get:

 1. xl vcpu-pin 3 all 4-6 : hard affinity for all dom3 vcpus --> 4-6
 2. xl vcpu-pin -s 3 all 3,5 : soft affinity for all dom3 vcpus --> 3,5
 3. xl vcpu-pin 3 all 4-6 3,5 : combination of 1 and 2
 4. xl vcpu-pin 3 all - 3,5 : same as 2

Would that be ok?

Thanks and Regards,
Dario

-- 
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://about.me/dario.faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)

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

_______________________________________________
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®.