On Tuesday, 09/26/2006 at 03:25 AST, "Scully, William P" 
<[EMAIL PROTECTED]> wrote:
> Let me refine this question.
> 
> - Two users.  Similar workload.  A real processor with multiple CPUs.
> The underlying CP has no resource constraints.  CP SHARE the same for
> both users.  The only "difference" is user A has one virtual processor
> and user B has two virtual processors.
> 
> I'm not really trying to solve a problem by asking this question.  Just
> trying to understand how CP tries to give everyone "their share".  I
> think we all expect that two users, over time, doing the same work, with
> the same virtual machine settings, get about the same resources from CP.
> I'm asking, does giving a user an extra virtual CPU imply that that user
> is going to get more CPU cycles, merely because of the second virtual
> CPU?

No.  If the virtual machine cannot make use of the 2nd CPU it will, in 
fact, only get up to *half* of the CPU resources you allocated to it. When 
you add virtual CPUs, you enable the virtual machine to overlap more 
operations and run multiple threads concurrently.  This can give the 
*appearance* of more horsepower but is really just loosening the 
bottleneck of "access to a CPU".  The virtual machine gets more done in 
the same unit of wall clock time.

If the virtual machine spends most of its time waiting for humans or 
devices, then nothing will help since the CPU instructions aren't the 
limiting resource.
 
> I see Marty thinks the answer is no.  That's my feeling too.  Did this
> change at z/VM vs. VM/ESA?  Or was it always this way?

It's been this way forever.

Alan Altmark
z/VM Development
IBM Endicott

Reply via email to