[Haizea] Multi-instance restype vs. higher capacity

[Shi Jin]

That's very interesting. In a virtualization environment, which scenario
makes more sense? If seems first that scenario (1) is natural but if the
hypervisor is  smart enough (e.g, with cgroups, maybe) scenario (2) seems
possible and makes better resource utilization.

Is there a way for haizea to choose to work with scenario (2) (I think the
current ONE+Haizea is scenario (1), right?).

Thanks.

Shi

On Fri, Oct 15, 2010 at 3:39 PM, Borja Sotomayor <borja at borjanet.com> wrote:

> Hi,
>
> > From scheduling point of view defining the following definitions are the
> same?
> >
> > 1) Defining a cpu resource as 8 instance (within multi-instance
> > restype) with capacity 100
> > 2) Defining a cpu resource as single instance with capacity 800
>
> If you assume that leases always request entire CPUs, yes.
>
> Otherwise, no. Let's suppose it's 2-instance and 200. If you get three
> lease requests, one needing 50% of the CPU and the other two needing
> 75% of the CPU, in scenario (1) above, Haizea would be unable to
> schedule them into the two CPUs, but *would* be able to schedule them
> in scenario (2).
>
> Cheers!
> --
> Borja Sotomayor
> -Scientific Writer at UChicago's Computation Institute
> http://people.cs.uchicago.edu/~borja/
> -Community Manager for the OpenNebula project
> http://www.opennebula.org/
> _______________________________________________
> Haizea mailing list
> Haizea at mailman.cs.uchicago.edu
> https://mailman.cs.uchicago.edu/mailman/listinfo/haizea
>



-- 
Shi Jin, Ph.D.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/haizea/attachments/20101015/cc868de4/attachment.htm