The big advantage of VM nowadays is the sharing of real memory. LPARs can
 share processor and 
channels and devices. Minidisks are pretty small, for z/OS there is usual
ly no particular advantage 
to them.

So the question is, do you have workloads that can take advantage of shar
ed memory? Production 
z/OS systems with different memory peaks might be complimentary. Test sys
tems that are only up 
part of the time are an even better candidate. If all your z/OS systems h
ave peak storage 
requirements at the same time, then memory savings will be limited.

You can  overcommit memory -- but the bad news is that when the z/OS gues
t takes a page fault, 
the whole guest stops until the page fault is satisfied. VSE has implemen
ted handshaking so that 
only the particular task whose page is faulted has to stop, but so far z/
OS has never supported 
that. So for production z/OS guests, you really don't want to take page f
aults. For test z/OS  
guests this may be tolerable.

There is a CPU cost to running under z/VM, and reliability will be slight
ly lower (?), so you have a 
trade off between storage savings and other overhead. 

Also, there is the problem of having yet another operating system, with d
ifferent commands and 
different command syntax, for your operations staff to support. If they h
ave no VM experience, 
then they are likely to oppose you. People who are used to all those comm
as get very upset that 
VM does not use them! Tuning another operating system may also be a probl
em for your 
performance or capacity planning people.

We use z/VM to run large numbers of DR and system test z/OS guests. We do
 not run z/OS 
production that way. (It somewhat depends on whether you consider DR to b
e production or test.) 
Generally, each group (for example the IMS group, the DB2 group, the z/OS
 group, etc.) has its 
own set of z/OS test guests. This has the big advantage that when the pho
ne rings and the system 
programmer has to "leave" to fix a production problem, he (or she) doesn'
t lose his (or her) test 
shot -- when they come back, the guest is waiting where they left off. Mo
st of the pages are 
paged out, so they don't cost much. (Even an "idle" z/OS guest  consumes 
cycles, alas.) 

Similarly we have DR guests corresponding to real z/OS LPARS. We use Capa
city On Demand to 
add processors when we are going to run a DR test with large numbers of g
uests all active at the 
same time. 

Another advantage of using VM is that the systems programmers can perform
 their tests 
(including "destructive testing") during regular business hours, instead 
of having to come in nights 
and weekends for test shots. That saves us quite a bit of overtime. Our D
R tests are still run 
mostly on weekends (the limitation is the number of tape drives, not CPU 
cycles).

Reply via email to