mcguire at neurotica.com
Fri Nov 30 13:20:50 EST 2018
On 11/30/18 7:09 AM, Dean Notarnicola via vcf-midatlantic wrote:
> I'm not going to argue with anyone's points here, as they are all valid.
> I've also been in IT for 35+ years and have built and maintained days
> centers. The only thing I will contradict is the idea that cloud is just
> "someone else's computer." Yes, they belong to someone else.
Not to be argumentative, but I will emphasize that this was the point
I was making. Those computers do, in fact, belong to someone else.
They are sitting in a building(s) that belong to someone else. And
someone else controls them, and someone else is responsible for their
maintenance, in every respect.
> However, cloud (by definition) is not timesharing, it is not co-location.
I agree that it's not co-location, but I'd ask you to tell me how,
exactly, it isn't timesharing. Elastic, capacity-on-demand
functionality is typically implemented using virtual machines, with
behind-the-scenes migration of VMs from host to host, in response to
hardware failures, reorganization, or just plain moment-to-moment
operating conditions. We all know that...and I know YOU know that; I'm
not "talking down" to you here. My point is, your VMs are sharing a
machine, or a set of machines, and mass storage, in the larger sense.
It's not likely that any (smaller) customer of such a service provider
will have any visibility into exactly where their VM lives and what else
is running on that processor core at any given time.
This is, by definition, timesharing. If you disagree, I'd love to
hear your point of view as to why.
Now, if you are a big enough customer to have a "nobody else on this
hardware" deal with the provider, that's great...and moves it a big step
closer to being co-location. Admittedly though, not completely so, but
I couldn't not point it out. ;)
> One of the
> benefits of cloud computing is to offer nearly instant elastic, on demand
> resources for variable workloads, such as software development, seasonal or
> other intermittent tasks. Traditionally, we would overbuild our data center
> to handle these things and as a result, 80% of our resources would be at
> 10% to 20% utilization for the majority of the year, wasting energy, space
> and the time of our staff to maintain them. To be sure, we keep core,
> business critical systems on prem (with HA and DR in co-lo) but that's ~5%
> of our infrastructure. After extensive risk analysis, we are able to put
> everything else in cloud, with the proper security and availability
> controls that meet our needs (we are a pharmaceutical company and most
> maintain GxP environments.) As a result, we only pay for the resources we
> actually utilize, can allocate and de-allocate resources on demand. As
> well, our lean IT staff can concentrate on higher value work.
The benefits are huge, only a moron would dispute that. It's
incredible, the things a modern VM-based hosting network can do, and the
benefits to their customers are very real and absolutely huge.
But those benefits do not come without a cost: Direct control over
just about everything is lost. Some suitly types may say that this is a
good thing, but I personally would never accept that. That is one of
the reasons I left the business of "big I.T." Decisions are now being
made exclusively for suitly reasons, not technical reasons. We all
remember the beginnings of this, when suits in charge of I.T. started
valuing contracts over OS source code licenses. They don't want ways to
fix things quickly, they want someone to scream at when it breaks. I'm
a technical person; I don't work that way.
You said above that your core business-critical systems are kept
in-house, and for you, that's about 5% of your infrastructure. In my
case, my core business-critical systems are about 80% of my
infrastructure. In other words, my mileage did, in fact, vary.
For that other 20%, I'm envisioning of all the idiots I've interviewed
and rejected in the past, who went on to work for Rackspace...I'll keep
that 20% in-house, thanks. ;)
Further, benefits or not, it's still "someone else's computer".
> To be sure, cloud is not a magic bullet. Getting true value from it
> requires due diligence in knowing your compute environment, your business
> and security needs, and creating/maintaining proper governance and
> controls. It may not be cheaper, but done properly can be more efficient
> and result in a much more agile infrastructure. Again, YMMV.
Agreed on all points. But VCF is neither Netflix, AirBNB, Atlassian,
nor a pharmaceutical company. The moment VCF's internal operations
begin to depend on an agile infrastructure with capacity-on-demand etc,
I'm sure things will be looked at a bit differently.
> Tread lightly :-)
You've met me, right? ;)
Dave McGuire, AK4HZ
New Kensington, PA
More information about the vcf-midatlantic