Moving z/OS from LPARs to z/VM guests

[Bill Bitner]

Brian,
A few urls that might help, some are very old but might jar other thoughts:
http://www.vm.ibm.com/perf/tips/z890.html
http://www.vm.ibm.com/perf/tips/prgvse.html
http://www.vm.ibm.com/perf/tips/2gstorag.html
http://www.vm.ibm.com/perf/tips/os390per.html

One does need to be careful with some of the SMF records. I do not have
my list handy, but there are some that would need to be adjusted and
others that are just fine when z/OS is on VM. Which ones do you typically
use?

Bill Bitner - VM Performance Evaluation - IBM Endicott - 607-429-3286

[Bill Bitner]

The most simular thing to hypersockets between LPARs is Guest Lans
between machines. It performs much better than hypersockets. Virtual
Switch is a special case of Guest Lans where the Guest Lan is connected
to the OSA. There are performance reasons to use Guest Lans over the
Virtual Switch.

There is Guest LAN HiperSockets if you want to match the HiperSockets of
LPAR for function; or the real HiperSockets can be virtualized.

One of the reasons to leave a large system, such as z/OS in its own
LPAR, was eliminated with z/VM 5.2. The reason had to do with z/VM 5.1
and under, limitation of 2 GB of central storage. The rest was expanded
storage. That is old hat now. If you hear others bashing z/VM over
this issue, as long as your processor can run z/VM 5.2....old news.

To be clear here, in many cases you can use more than 2GB of central
storage. I believe the point Tom is making is that, prior to V5.2, there
were high requirements for memory below 2GB. See
http://www.vm.ibm.com/perf/tips/2gstorag.html for more details.

Bill Bitner - VM Performance Evaluation - IBM Endicott - 607-429-3286

[Brian Nielsen]

I obtained a report that shows the number & type of SMF records generated
on z/OS last month. There are 53 types with non-zero counts. If the list
of ones to be careful with is around, and what adjustments need to be
made, I'd like to see it.

Brian Nielsen

Post by Bill Bitner
One does need to be careful with some of the SMF records. I do not have
my list handy, but there are some that would need to be adjusted and
others that are just fine when z/OS is on VM. Which ones do you typically
use?
Bill Bitner - VM Performance Evaluation - IBM Endicott - 607-429-3286
=========================================================================

[Brian Nielsen]

On Fri, 10 Feb 2006 17:03:01 -0600, Brian Nielsen

Many (many) years ago I used to run MVS under VM/SP on a 4381, so that
environment isn't new to me, it's just not recent vintage.

Just for reference, this is one of the earliest times I had MVS under VM
(I'd have to check my notes to see if we had it that way on the prior
4341). Many upgrades later it was MVS under VM/ESA on a 3090, but that
was still 10 years ago in a different shop.

Brian Nielsen

[Brian Nielsen]

I've read the referenced manual, and the SCRT tool appears to deal only
with the reporting aspect, and while that's important, it's not my main
concern.

I've read enough other related material that perhaps I can phrase my
question with more precision and correct terms.

With z/OS running directly in an LPAR: When the four-hour rolling average
MSU consumption is greater than the defined Capacity MSUs, the LPAR gets
soft-capped via Workload Manager interacting with PR/SM.

If z/OS is a guest of z/VM: It seems inappropriate, and perhaps not
possible, for WLM in z/OS to implement soft-capping of the z/VM LPAR. It
seems more appropriate for some WLM equivalent software at the z/VM level
to implement soft-capping of the z/VM LPAR.

What are the options (if any) in order to keep a z/VM LPAR within some
defined Capacity MSU's the way WLM does for a z/OS LPAR?

Brian Nielsen


On Mon, 13 Feb 2006 14:21:41 -0500, David J. Chase

There are some items I don't know enough about yet to gauge the
- Workload Manager is used to throttle back the z/OS LPARS below a
specified 4 hour rolling average of CPU usage (for cost reasons).
I've never used Workload Manager, but wonder: (a) will it work if
z/OS is a guest of z/VM, and (b) if not, what would accomplish the
same thing? Setting SHAREs is obviously not up to the task because
we're talking about the whole CP side.

Unless all of the z/OS images on the machine (guest or LPAR) are at the
z/OS 1.7 level then running any z/OS guest under z/VM will cause the
software charges for the z/OS guests to be levied against the size of
the VM LPAR and not against the 4 hour rolling average utilization.
Please see Chapter 5 of SG24-6522-16 Using the Sub-Capacity Reporting
Tool (aka the SCRT User's Guide)for more information.
David
-- David J. Chase, WW zSeries Software Sales --
-- IBM 18th Fl, 11 Madison Ave, NYC, NY 10010 --
=========================================================================

[Stephen Frazier]

Are you wanting to cap z/VM or the z/OS guest(s) running on z/VM?

Post by Brian Nielsen
I've read the referenced manual, and the SCRT tool appears to deal only
with the reporting aspect, and while that's important, it's not my main
concern.
I've read enough other related material that perhaps I can phrase my
question with more precision and correct terms.
With z/OS running directly in an LPAR: When the four-hour rolling average
MSU consumption is greater than the defined Capacity MSUs, the LPAR gets
soft-capped via Workload Manager interacting with PR/SM.
If z/OS is a guest of z/VM: It seems inappropriate, and perhaps not
possible, for WLM in z/OS to implement soft-capping of the z/VM LPAR. It
seems more appropriate for some WLM equivalent software at the z/VM level
to implement soft-capping of the z/VM LPAR.
What are the options (if any) in order to keep a z/VM LPAR within some
defined Capacity MSU's the way WLM does for a z/OS LPAR?
Brian Nielsen

--
Stephen Frazier
Information Technology Unit
Oklahoma Department of Corrections
3400 Martin Luther King
Oklahoma City, Ok, 73111-4298
Tel.: [phone]
Fax: [phone]
Pager: [phone]
email: stevef%doc.state.ok.us

[Brian Nielsen]

Probably both.

The z/OS's under z/VM need capping for purposes of billing the software
that runs in each z/OS image. (Can WLM in z/OS, or some other software at
the z/VM level, adjust an ABSOLUTE SHARE for the z/OS guests?)

I'm not sure on the billing for z/VM if there is an option for the 4-hour
rolling average MSU consumption method. If so, that would require some
WLM equivalent on z/VM to implement soft-capping for the z/VM LPAR.

Brian Nielsen

On Thu, 16 Feb 2006 16:48:27 -0600, Stephen Frazier

Post by Stephen Frazier
Are you wanting to cap z/VM or the z/OS guest(s) running on z/VM?

Post by Brian Nielsen
I've read the referenced manual, and the SCRT tool appears to deal only
with the reporting aspect, and while that's important, it's not my main
concern.
I've read enough other related material that perhaps I can phrase my
question with more precision and correct terms.
With z/OS running directly in an LPAR: When the four-hour rolling

average

Post by Stephen Frazier

Post by Brian Nielsen
MSU consumption is greater than the defined Capacity MSUs, the LPAR gets
soft-capped via Workload Manager interacting with PR/SM.
If z/OS is a guest of z/VM: It seems inappropriate, and perhaps not
possible, for WLM in z/OS to implement soft-capping of the z/VM LPAR.

It

Post by Stephen Frazier

Post by Brian Nielsen
seems more appropriate for some WLM equivalent software at the z/VM

level

Post by Stephen Frazier

Post by Brian Nielsen
to implement soft-capping of the z/VM LPAR.
What are the options (if any) in order to keep a z/VM LPAR within some
defined Capacity MSU's the way WLM does for a z/OS LPAR?
Brian Nielsen

--
Stephen Frazier
Information Technology Unit
Oklahoma Department of Corrections
3400 Martin Luther King
Oklahoma City, Ok, 73111-4298
Tel.: [phone]
Fax: [phone]
Pager: [phone]
email: stevef%doc.state.ok.us
=========================================================================

[David Boyes]

In theory you can post-process the SCRT reports and
dynamically adjust z/OS's SHARE setting. If you run an z/OS
program in supverisor state, you can issue diagnose 8 from
z/OS. Diagnose 8 is how programs issue CP commands like "MSG
ALAN AVGMSU = 483". Then user ALAN can trap that and adjust
the SHARE up or down as needed, re-evaluating it every, say,
30 minutes.

Sounds like a useful improvement to z/OS. It'd be nice if they learned
about closing spool files properly too.

I can see that having VMRM perform group-level calculations
based on MSUs reported by z/OS would be a useful function.

It'd be even more useful if it didn't require z/OS, or required at worst
z/OS.e. This is kind of the next-order scheduling function -- you'll
need it for fleets of zLinux systems too. Maybe a z/OS appliance...hmm.

[Brian Nielsen]

z/VM cannot restrict a guest based on MSU consumption, nor can WLM in z/OS
manipulate the SHARE given to a guest. The most WLM can do is adjust the
entire z/VM LPAR's weight.

What privileges/options does a z/OS guest need to be able to adjust the
z/VM LPAR's weight?

When hosting other peoples z/OS images it would be important to prevent
them from doing so if they tried. The thought of having multiple z/OS
guests fighting over the LPAR weight is even worse.

It seems better to run SCRT
in the guests and adjust the SHARE of the z/OS guests so that their
highest reported rolling 4-hour average consumption doesn't exceed
whatever value you want.

So, if I understand you correctly:
1) There's nothing that does that now for guests.
and,
2) There's nothing at the z/VM level to manage the rolling 4-hour
average of the whole LPAR.


Brian Nielsen