[Info-vax] OpenVMS servers and clusters as a cloud service

[Stephen Hoffman]

On 2018-01-01 13:07:26 +0000, Scott Dorsey said:

> Kerry Main <kemain.nospam at gmail.com> wrote:
>> 
>> My point is that I know Amazon and Google etc. have very high numbers 
>> of servers (phys or VM really does not really matter from admin 
>> perspective), but their Sysadmin ratio is not 28,000 servers for a = 
>> single admin. Hung services, hung/crashed servers, log monitoring, 
>> backup failures, password mgmt., hardware failures, firewall rules 
>> integration are all examples of Sysadmin activities where tools and 
>> custom automation can certainly help.
> 
> No, that's high but not insane.  My wife works in a place where it's 
> about one admin per 5,000 machines.  If something goes wrong with a 
> server, they save the user partition, wipe the machine, replace the OS 
> image, put the user partition back in about five minutes.  They don't 
> diagnose things or repair them in most cases, because the VM technology 
> has made it so easy to replace them.
> 
> They have a small team that makes standard images, but the day to day 
> admin work is almost entirely automated.  They have six Linux admins 
> for the whole  multinational corporation, spaced around the world so 
> there are always two people on duty on any given shift.
> 
> It's a very different philosophy than we came up with.  Hell, I worked 
> for  a company where they probably had 300 people on staff supporting 
> one machine, way back when.

That's a direct evolution from what even DEC was doing with mass 
deployments, back in the day.   Only a few folks for an entire office 
campus, and with related support staff increasingly working remote and 
shared across sites even back then.

Biggest local shift is a move away from using imaging for new client 
and server installs, and toward automated software installations.  Base 
system install gets loaded — and interestingly, the operating system 
equivalents of SYS$SYSTEM and SYS$LIBRARY/SYS$SHARE and SYS$UPDATE 
directories are all locked down and unmodifiable in the typical 
install, save for vendor-provided updates — and the necessary app 
bundles and local certificates and app profiles are then loaded atop 
that.   Layering.  The system vendor is also doing the core 
provisioning for new systems.  Not the local IT folks and the end-users 
and any ISV that might be involved.  Like many system vendors, they've 
been shipping factory-installed software for a very long time, and now 
with the system serial numbers keyed to the organization in the 
enrollment servers and the provisioning servers, meaning that there's 
no need to physically access even brand-new systems destined for racks 
or for end-users.   The OpenVMS Factory Installed Software approach 
isn't even a fraction of what's involved here, either.

So...  folks using OpenVMS do one or two or few installs on hosted 
environments to start out with and follow the traditional high-touch 
approach, as VSI gets the port sorted out and works out their policies 
and procedures and their implementations around installations and 
console access and licensing for remote deployments.  Give it a few 
more years before we see locked-down system software installations, 
re-worked app installations, and apps that don't need to spray bits all 
over the system disk for basic functions like starting up and shutting 
down; the tools for larger and faster deployments and redeployments, 
and with less interaction.

Past the scale of SOHO and hobbyist deployments, servers are 
plug-replaceable, and not bespoke.  Livestock.  Not pets.  And whether 
the servers are local or are hosted, the basic issues and requirements 
are very similar.

I'd prefer to not get wrapped up with what server folks are doing now, 
look at where they're likely headed and what problems they'll encounter 
on the way, and definitely don't get tied to what approaches are used 
now or what works now, as that's all history.  Sizes and scale are 
certainly useful data, but it's the trends in numbers of staff and 
servers and automation that are what any vendor is looking at.   ISVs 
and end-users?  Look around.  Learn what other folks are doing, too.

And as I've mentioned once or twice, design nostalgia isn't a viable 
long-term technical strategy for those designs that aren't meeting 
current requirements and expectations.



-- 
Pure Personal Opinion | HoffmanLabs LLC