I dare to disagree a bit:
- MWV is required to get Virtual Reserve/Release, that is sharing inside
z/VM.
- SET SHARED ON and FULLPACK Mdisk is for Real Reserve/Release

They are different beasts:
The Real R/R is a function of the dasd controller: when one system sends a
Reserve, it halts the IOs from other systems.  "System" was originally a
CPU, now it can, be an LPAR.  Multipathing since XA made it even more
complex.  One invented "path groups"; z/OS, z/VM etc send their unique path
group id to the controller and a reserve then locks out all other path
groups.

Real R/R could never help to protect virtual machines from each-other (their
IO come from the same path group ID), so CP has to come in and lock out IO's
from other guests when one guest issued a reserve.   That's Virtual R/R.
And luckily: both can be combined too.

I even once created a local mod to allow real R/R on non-fullpack minidisks,
at a time we had the requirement to share 8 different minidisks amongst VM
systems.  8 fullpacks was a bit too much in my eyes.  It was an easy mod,
but we had to assure not more than 1 minidisk with R/R requirement was on
these packs

2010/7/28 Alan Altmark <[email protected]>

> On Wednesday, 07/28/2010 at 10:03 EDT, Eginhard Jaeger
> <[email protected]> wrote:
> > Is that really true? When checking just now I found that the 'Planning
> > and Administration' manual also says so, but I'm 99% sure (allowing
> > 1% for memory deterioration) that I shared a 20 cyl. RACF database
> > between two separate VM XA systems and it worked perfectly well.
> > While the rest of the packs certainly shouldn't be used for any high
> > activity stuff I believe we did actually use it (space was still quite
> > expensive 20 years ago).
> > So I suspect that the 'full pack' requirement may just come from sharing
> > DASD with non-VM systems, and that it doesn't really apply to the
> > RACF data base being shared between multiple VM systems only.
> > But I never had to share disks for later VM releases and so don't
> > know for sure ..
>
> Yes, it's true.  CP cannot propagate a virtual Reserve/Release to the REAL
> dasd if the volume contains multiple minidisks.  That would cause users of
> minidisk A to be affected by a reserve on minidisk B.  (And reserves can
> be held permanently, if desired.)
>
> If propagation to the real dasd isn't required then CP can simply the
> reserves himself and it is ok to have multiple minidisks on the volume.
>
> Hence the rule for cross-system shared DASD that requires Reserve/Release:
> - Full pack minidisk only
> - Virtual reserve/release is enabled (MWV on the MDISK)
> - The RDEVICE is configured as SHARED in SYSTEM CONFIG
>
> If only one guest on the VM system needs the device, then you can dedicate
> it, but you don't want to do that with RACF since you may want to be
> testing something with RACMAINT while RACFVM is running, or you may want
> multiple RACFVMs.
>
> Since it's all just ones and zeroes, it is theoretically possible to
> invent a "1-END" minidisk that would cause a full-pack reserve to flow,
> but we're talking about RACF, a highly trusted component of the system, so
> I'm not really worried about something hacking in and doing violence to
> the volser.  "Acceptable risk."
>
> What I *am* looking at is finding a way to let RACF detect non-fullpack
> configurations when it is configured for cross-system sharing.  The
> default minidisk configurations work ok because they are defined MR, not
> MWV.  The lack of the "V" causes RACF's attempts to ENQ (reserve) the
> volume to fail.  That triggers RACF to cache database content for
> performance instead of doing I/O all the time, KNOWING that there is no
> sharing.  Changing to MWV is sufficient for adding extra RACF servers or
> z/OS guests on the same system, but not for cross-LPAR sharing.
> Unfortunately, RACF does not today require you to declare your intent vis
> a vis database sharing and so cannot enforce any particular configuration.
>
> Alan Altmark
> z/VM Development
> IBM Endicott
>



-- 
Kris Buelens,
IBM Belgium, VM customer support

Reply via email to