On Fri, 31 Oct 2008 11:37:40 -0400, Alan Altmark <[EMAIL PROTECTED]
> wrote:

>On Thursday, 10/30/2008 at 10:29 EDT, David Kreuter
>While I will grant you the "optimization" point, let's not get too carri
ed
>away.  In an LPAR, SIE handles guest I/O only for dedicated OSA and FCP
>adapters.  All other I/O is virtualized by CP.  SIE *does* handle CP's
>I/O!
>
>The emphasis on the use of a virtual switch rather than dedicated OSAs
>leaves us with primarily FCP adapters for SCSI.
>
>But with all that said, as others have pointed out, the word "overhead"
>has no meaning.  Yes, there is overhead and sometimes, yes, it can be "n
ot
>insignificant".  The question is whether the applications are meeting
>their SLAs and whether the IT provider is meeting its expense goals.  "C
an
>I get acceptable application performance at a cost I can afford?"  As wa
s
>mentioned, you may have more overhead handling an I/O request, but if yo
u
>can satisfy it from MDC, it was time well-spent.  Assuming, of course,
>you've got the CPU available to handle the I/O request!
>
>Alan Altmark
>z/VM Development
>IBM Endicott
>========================
=========================
==========
==============

The person who asked the question has been asked to review virtualization
 options (VMWARE, p-
series, and z/VM) and determine which workloads should go where. Since th
 midrange folks have 
so far decided "we don't need z/VM, we have VMWARE", this is actually pro
gress.

Unfortunately, his background is all in PCs and Linux (and i-series), no 
mainframe. But he really is 
trying to understand the mainframe. He wants to know if VM is like the VM
WARE he is failiar with -
-  at least he ASKED instaed of making assumptions.

I don't think that the word "overhead" has no meaning. I'm pretty sure he
 meant CPU overhead 
comparing native to a guest. I think that he has observed or been told th
at running Linux guests 
under VMWARE takes more processor than without VMWARE, and that this incr
eases for heavy I/O 
workloads. 

I would expect the same thing to be true for z/VM, but I would HOPE that 
the mainframe does it 
better. Almost all I/O on a PC is handled by the central processors, whil
e the mainframe has the 
IOPs to do much of that work. 

It would be nice to quantify this, though.

Alan Ackerman
Alan (dot) Ackerman (at) Bank of America (dot) com 

Reply via email to