Question concerning running z/OS LPARs under z/VM

[Mark Yuhas]

I was quesioned about reconfiguring our current processor.

Currently, we are running 6 LPARs - 2 sandbox LPARs , an applicaton/test LPAR, a certification LPAR, a production LPAR and a scheduling LPAR. The sandbox LPARs run under separate monoplexes and separate MASs. The remaining LPARs run under a Base Sysplex and the same MAS.

The configuration in question would have the 2 sandbox LPARs, the application/test LPAR and the certification LPAR run as guests under z/VM.
The production and scheduling LPARs would still run as separate LPARs but the same MAS.

Can the current Base Sysplex and MAS still be used with 2 members running native and the other two runnng as guests under z/VM? Would we need a second MAS?

I would also think that there would be extra overhead with PRSM deciding which LPAR gets resources and if it is the z/VM LPAR, then z/VM deciding which guest gets the resources. Seems like a duplication of effort. Then, of course, would JES2 tolerate this as well?

Any thoughts, or, has anyone performed this type of configuration.

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@listserv.ua.edu with the message: INFO IBM-MAIN

[Paul Gilmartin]

Post by Mark Yuhas
I would also think that there would be extra overhead with PRSM deciding which LPAR gets resources and if it is the z/VM LPAR, then z/VM deciding which guest gets the resources. Seems like a duplication of effort. Then, of course, would JES2 tolerate this as well?

Perhaps even worse, nested paging. Might that be alleviated by either
some use of V=R (is this possible?) so only the guests but not CP pages,
or by defining an absurdly large virtual storage for the guests so only
CP but not the guests page?

-- gil

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@listserv.ua.edu with the message: INFO IBM-MAIN

[Staller, Allan]

z/VM solved the nested paging issue the late 80's. Usually by making the guest image so large that paging never became an issue.
There was also the "preferred guest" feature.


<snip>

Post by Mark Yuhas
I would also think that there would be extra overhead with PRSM deciding which LPAR gets resources and if it is the z/VM LPAR, then z/VM deciding which guest gets the resources. Seems like a duplication of effort. Then, of course, would JES2 tolerate this as well?

Perhaps even worse, nested paging. Might that be alleviated by either some use of V=R (is this possible?) so only the guests but not CP pages, or by defining an absurdly large virtual storage for the guests so only CP but not the guests page?
</snip>

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@listserv.ua.edu with the message: INFO IBM-MAIN

[Anne & Lynn Wheeler]

Post by Staller, Allan
z/VM solved the nested paging issue the late 80's. Usually by making
the guest image so large that paging never became an issue.
There was also the "preferred guest" feature.

The problem was worse than you can ever imagine. I pontificated a lot
about this in the mid-70s

I had done a lot of paging algorithm stuff as undergraduate in the 60s
... and got into tiff with the POK crowd when they started adding paging
to MVT (something they thought was benefit ... was actually picking
shared, high-use linkpac pages for replacement before lower-use private
data pages ... which didn't get fixed until well into the MVS release
cycle).

The other problem was running under vm370 ... most page replacement
algorithms assume some flavor of LRU ... least recently used ... the
page that has been least recently used in the past is assumed to have
the lowest probability of use in the future. When MVS is running some
flavor of LRU ... it is looking for the page with very low usage to use
its real storage location.

when running under vm370, with both vm370 and mvs running some flavor of
LRU ... the apparent page use behavior of MVS running in vm370 virtual
machine ... is the least recently used page (from vm370 point-of-view)
is the most likely to be used next (not least likely) ... because MVS is
out looking for the leased recently used for its next likely to be used
aka a LRU algorithm running under a LRU algorithm starts to look a lot
like an MRU (the least recently used page is the most likely to be
used).

This also comes up with large DBMS record caches which also tend to be
managed with least-recently-used replacement strategies ... and why
large DBMS record caches also tend to have to be PIN'ed (because the
operating system will have a tendency to replace the least recently used
page ... which has high probability of being needed by the DBMS).

misc. past posts about page replacement algorithms
http://www.garlic.com/~lynn/subtopic.html#clock

[shmuel+ (Shmuel Metz , Seymour J.)]

In <***@listserv.ua.edu>, on
09/15/2014
at 01:54 PM, Paul Gilmartin

Post by Paul Gilmartin
Perhaps even worse, nested paging.

AFAIK, PR/SM is strictly V=F, so no paging.

[Staller, Allan]

I do not see any gain from the proposed configuration, however, my thoughts are:

As I interpret the proposed configuration, you would be processing a total of 3 lpars (production, scheduling, and z/VM).
I do not see any unsolvable issues of a configuration nature, with sharing between z/OS guest images and non-guest images.

I do, however, see potential performance issues (all solvable) within that configuration.

What about z/VM maintenance? Think about a z/VM sandbox LPAR. You did not specify what level of z/VM is to be used.
If z/VM with Single System Image (?) is installed, z/VM allows live guest migration. i.e. guests can be moved from One VM image to another without interruption.
IIRC this is available w/z/VM 6.3 and later. Certain conditions apply. YMMV.

HTH,

Disclaimer. I am *NOT* a z/VM expert, although I have some familiarity with the capabilities. Your IBM/IBM partner tech rep should be able to correct anything above that is incorrect.


<snip>
Currently, we are running 6 LPARs - 2 sandbox LPARs , an applicaton/test LPAR, a certification LPAR, a production LPAR and a scheduling LPAR. The sandbox LPARs run under separate monoplexes and separate MASs. The remaining LPARs run under a Base Sysplex and the same MAS.

The configuration in question would have the 2 sandbox LPARs, the application/test LPAR and the certification LPAR run as guests under z/VM.
The production and scheduling LPARs would still run as separate LPARs but the same MAS.

Can the current Base Sysplex and MAS still be used with 2 members running native and the other two runnng as guests under z/VM? Would we need a second MAS?

I would also think that there would be extra overhead with PRSM deciding which LPAR gets resources and if it is the z/VM LPAR, then z/VM deciding which guest gets the resources. Seems like a duplication of effort. Then, of course, would JES2 tolerate this as well?
</snip>

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@listserv.ua.edu with the message: INFO IBM-MAIN

[M***@pi-sysprog.de]

I see no problem with the proposed configuration.

just a few thoughts:

Sharing resources becomes easier under VM from guest to guest (as opposed to systems from one LPAR to another)

Live relocation is not available to op-sys like z/VSE or z/OS (as of now).

Performance is no longer (more than 15 years) a problem (unless you are native already at 95%) - since the 1 MIPS 370/158 we have seen an awefull lot of improvements to guest performance - (just one: SIE)

With a z/VM on the CEC you could start exploring options with linux on Z

Martin

[shmuel+ (Shmuel Metz , Seymour J.)]

In
<***@ITDRENMX05.na.paccar.com>,
on 09/15/2014
at 06:34 PM, Mark Yuhas

Post by Mark Yuhas
Can the current Base Sysplex and MAS still be used with 2 members
running native and the other two runnng as guests under z/VM?

Why not, as long as you preserve the ETR/STP and CTCA connections?

Post by Mark Yuhas
Would we need a second MAS?

I see no reason why.

Post by Mark Yuhas
Any thoughts,

Do you expect to stay on Base Sysplex forever?

[Dana Mitchell]

Post by Mark Yuhas
Can the current Base Sysplex and MAS still be used with 2 members running native and the other two runnng as guests under z/VM? Would we need a second MAS?

As others have said, there is really no reason this couldn't be made to work, as long as all neccessary CTC's and timer links were available. I'm just wondering what problem you are trying to solve by doing this?

Dana

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@listserv.ua.edu with the message: INFO IBM-MAIN

[Paul Gilmartin]

Post by shmuel+ (Shmuel Metz , Seymour J.)

Post by Paul Gilmartin
Perhaps even worse, nested paging.

AFAIK, PR/SM is strictly V=F, so no paging.

And how is that relevant to the OP's question about VM guests?

(Well, "z/OS LPARs under z/VM" is sort of a contradiction. But the
OP said he was considering converting two LPARs to VM guests.)

-- gil

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to ***@listserv.ua.edu with the message: INFO IBM-MAIN