[Info-vax] VMware

[Arne Vajhøj]

On 12/10/2019 7:05 AM, Bob Gezelter wrote:
> On Tuesday, December 10, 2019 at 4:26:29 AM UTC-5, Ian Miller wrote:
>> On Tuesday, December 10, 2019 at 3:22:51 AM UTC, Dave Froble
>> wrote:
>>> But, I have to ask, how do they run all those instances.  It
>>> seems to me that it would be an operations nightmare.  I'm
>>> guessing they use SANs so backup would not be such an issue.  But
>>> other operations?

>> Hi David, to run lots of VM systems requires massive automation
>> otherwise they have to employ lots of sysadmins and that costs to
>> much. With the automation you can create a VM with linux+oracle etc
>> at the press of a button in a very short time. All the VMs are
>> monitored automatically, some fixes for problems are applied
>> automatically and so on. Running 10,000 systems in a data center
>> with a few people is possible.
>> 
>> It is a strange new industrial big scale world. I mostly still
>> handcraft VMS clusters from the finest ingredients so am a niche
>> craftsman sitting in a corner while the young folk play with the
>> shiny new industrial automated sysadmin toys.

> Perhaps it is a more subtle point. It is not so much a new world, as
> a world with a broader spectrum of possible choices and wider set of
> alternatives.
> 
> In the Windows/Linux world, many applications presume that they are
> on a dedicated instance. Attempting to run multiple applications in a
> single instance results in collisions, e.g., port numbers,
> serializations. Resolving such problems can be complex and
> time-consuming and requires the cooperation and assistance of
> outsiders. Using dedicated instances somewhat nullifies the problem
> by giving each their own "playpen" to operate within, which removes
> the issue of how well applications co-exist.
 >
 > It is also possible to have similar problems with databases and other
 > multi-client packages. Not everyone operates well in a shared
 > namespace.

Most of these problems are really OS agnostic.

But the handling of the problems are different:

expensive physical box and expensive OS => you work out the configuration

virtual environment and cheap OS => you just spin up a new VM and 5 
minutes after everything is working (less work than finding the manual 
to figure out how to change the default config)

Maybe slightly exaggerated. But the point is that one app per VM
philosophy is not because technically it has to be that way - it
is is just because it is cheaper that way.

> Testing is yet a third situation. Many times, I have needed to do an
> experiment with potential consequences. Better to do it on a
> disposable virtual instance (an approach I described in a blog
> article several years ago) than impact a longer lived instance. One
> starts with a disposable instance, and graduates to a less disposable
> environment after one has proven feasibility.
> 
> Similar arguments hold for training and proficiency. Crashing an
> inexpensive disposable instance is far less expensive than real
> hardware.

Yep.

Arne