Re: [Xen-devel] Re: Xen cpu Scheduling

[pak333@xxxxxxxxxxx]

Thanks for the explanation.

I still have questions about how cpu utilization is reported by xm top/xentop. If someone can answer the question as to how the cpu% is calculated, when a VM with 2 vcpus, is only using 1 pcpu at any given point of time.

 

Also, a vcpu is given the full pcpu. What if it doesn't use the full pcpu. How is it scheduled?

thanks

Prabha

 

 

> -------------- Original message -------------- 
> From: "Mike D. Day" <ncmike@xxxxxxxxxx> 
>
> > On 20/03/07 01:37 +0000, pak333@xxxxxxxxxxx wrote: 
> > > 
> > > 1. Suppose I give a VM 2 vcpus. When this VM gets scheduled will it 
> > > always be given two pcpus or can it be given 1vcpu also sometimes. 
> > > For example if we have mutiple VMs running, and if a cpu gets freed up 
> > > will my 2vcpu VM be scheduled on this single free cpu. If a single 
> > > cpu cna be given how is the cpu % calculated by xentop 
> > 
> > It depends on how busy the CPUs are on the machine, and the ratio of 
> > VCPU to PCPU. 
> > 
> > When the VM has two VCPUs, each VCPU (when runnable) will be put on 
> > the run queue of a physical CPU. This VM will run concurrently on AT 
> > MOST 2 PCPUs, sometimes only 1 PCPU. If on the platf
> orm VCPU == PCPU 
> > in aggregate then the VM with 2 VCPUs will frequently run concurrently 
> > on two PCPUs. 
> > 
> > 
> > > 
> > > 2. If the 2vcpu is scheduled on two cpus, and one of its threads goes 
> > > to sleep can that single thread be context swtiched out. or will it be 
> > > kept spinning. 
> > 
> > VM's get scheduled as VCPUs, not threads. Having two VCPU's for a VM 
> > is sort of like a process having two threads. But there are a lot of 
> > differences as well so don't take the analogy too far. 
> > 
> > When a VCPU for a VM is not running it is not executing any code. The 
> > VCPU can be runnable, running, or sleeping. It only executes code when 
> > it is running. 
> > 
> > However, remember that *kernel threads* are scheduled by the guest kernel, 
> > while the guest kernel is scheduled as one or more VCPUs by *xen.* The 
> > guest kernel is allowed to put its own thre
> ads to sleep and awaken 
> > other threads, but when Xen decides that the time slice is over for a 
> > VCPU, then that VCPU gets rescheduled at the back of the q, or put to 
> > sleep. 
> > 
> > I don't know very much about xentop. 
> > 
> > Mike 
> > 
> > -- 
> > Mike D. Day 
> > IBM LTC 
> > Cell: 919 412-3900 
> > Sametime: ncmike@xxxxxxxxxx AIM: ncmikeday Yahoo: ultra.runner 
> > PGP key: http://www.ncultra.org/ncmike/pubkey.asc 
> > 
> > _______________________________________________ 
> > Xen-devel mailing list 
> > Xen-devel@xxxxxxxxxxxxxxxxxxx 
> > http://lists.xensource.com/xen-devel 
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel