The Cranky Admin

Public Cloud Is Not a Utility

There will always be a role for on-premises datacenters.

There is a dangerous idea becoming very popular in technology circles. The idea is "computers are like electricity." This is the beginning of a line of reasoning that ends with all of humanity switching to a utility-style IT model and giving up our on-premises technology. The argument is that this will occur "just like people gave up running their own generators back in the old days." We did no such thing.

Before we dive into the so-called "inevitability" of the public cloud, let's get the record straight on electricity generation. For one thing, there are a heck of a lot of electric generators still operating in the hands of private individuals, corporations and government entities not associated with main grid power generation.

In the developed world, most families own an electric generator, many own more than one. We call them cars. For the most part, we don't use our cars as generators directly, but we certainly use them in place of centralized electricity.

We could have designed our society's infrastructure such that all personal transportation was electric and ran off the grid over 100 years ago. Cars could have relied on overhead wires, much as buses in my hometown once did. The individual cars would have been significantly cheaper and easier to maintain, with fewer moving parts, and they'd last longer.

The problem was that neither the vehicle manufacturing industry nor the average people buying personal transportation wanted this. Vehicle manufacturers did not want to start us down a road towards everyone owning long-lasting, inexpensive and easy to maintain cars. And we didn't want to be restricted in when and where we could drive.

Even if we ignore who killed the grid-backed electric car and why, there are dedicated electric generators out there in the world, to be sure. I've yet to see a proper datacenter that relied entirely on the main grid. They're popular on leisure properties, in the resource extraction industry and there is increasing use of microgeneration such as rooftop solar even in the developed world where national grids are a thing.

Let's also consider the significant portions of the world that rely on private generation of electricity because the main grid is unreliable, nonexistent, too expensive, or corrupt due to lack of regulation.  This idea that we collectively gave up on generating our own power and simply trusted in "the man" to deliver all our electrical requirements is a myth.

Where and when something really matters -- from standing up datacenters to ensuring our ability to commute to work -- those with the means overwhelmingly choose not to rely exclusively on centrally provided services.

It's All About the Use Cases
Nobody cares if you miss your nightly game of sportsball on the television, or you can't brew your cup of afternoon tea -- not even you. If the mains go out and you lose access to your creature comforts, you'll grump and you'll grumble, but that's where it will end. Unless and until grid outages start dragging on into days -- in many cases it takes weeks -- the milled masses won't get off the couch and into the streets.

Even here in Canada, where power outages caused by winter storms can be life threatening, this holds true. The 2016-2017 winter storm season in Canada broke records all across the country. The east and west coasts in particular were hit hard, with protracted power outages occurring frequently, in some cases lasting days.

Governments did not fall. Citizens did not riot. They calmly made their way to emergency shelters, suffered through it, and then largely forgot about it. This is because mains power is -- in the short term, at least -- a luxury, not a necessity.

Oh, without the main power grid civilization as we know it would collapse. Virtually every creature comfort we enjoy relies on it. Our entire national logistics system relies on it. Even food distribution is predicated on distributors, retails and individuals all having ubiquitous access to refrigeration, something that wouldn't be realistically possible without main grid electricity.

The power grid has existed, and been more or less reliable, for so long that every aspect of our entire society has evolved around the idea that any mains outage will last no more than a few days.

But we don't bet lives on it.

Beep. Beep. Beep.
With the grid down, we still expect to be able to dial emergency services on our phones, have those phones connect to an operator, have an ambulance arrive, and have our lives saved at a fully-powered and operational hospital. We bet our lives on this, individually and collectively, every single day.

Telecommunications infrastructure, both wired and wireless; emergency response call centers; and hospitals are some examples of services so absolutely vital that we legally require they be prepared for failure of the national grid. Regulation and law are critical concepts here, and they must inform any discussion about the so-called inevitability of the public cloud.

The service-level agreements (SLAs) the public cloud providers claim mean absolutely nothing. If you're down during the wrong hours, your organization could lose critical customers, ultimately causing your business to fold. The wrong outage at the wrong time, of the wrong length, can spell doom, even for the largest organizations.

A public cloud provider isn't going to pay you back the millions (or billions!) lost during an outage or a data loss event. You'll just have to accept a few credited hours and an admonishment that, were you doing things properly, you wouldn't have trusted everything to whatever configuration you created and instead used some other configuration that was more robust (and would have cost significantly more).

Backups matter. You must either have a copy of your workloads and data ready to go on another cloud provider, or an on-premises solution. The public cloud emphatically is not guaranteed uptime, it isn't "fire and forget," it isn't "always on," and it won't always be there. You can't bet lives on the public cloud any more than you can on mains power, and you shouldn't be betting livelihoods on it either.

Just as with the main power grid, when it really matters, best be sure you have a backup.

Perhaps the biggest problem with the "utility computing" concept is that public cloud providers aren't regulated like utilities. Power, water, sewage and natural gas (heat: very important in Canada), are all regulated.

Enron was a scandal because the utility provider decided to play silly buggers with power availability in order to jack up rates once they had a monopoly. Outside of antitrust laws, which don't seem to be enforced on tech companies in any meaningful manner and take a decade or more to litigate, there are no laws to prevent public cloud companies from doing whatever they want with their offering.

They could implement Uber-style "surge pricing" tomorrow then play with availability to drive up revenues. Once they're down to an oligopoly with on-premises hardware sales virtually non-existent, I'd bet significant sums of money that this is exactly what they'll do. And that's the most benign of the problems customers face.

With no regulation, there is no incentive except competition to ensure that maintenance is performed at levels adequate to ensure availability. Tech companies are very good at eliminating competition.

There is no incentive -- including competition -- for public cloud providers  to improve uptime, pressure Internet providers to ensure higher quality and more robust last-mile networks, or otherwise make their solutions as reliable as mains power. Existing legal frameworks allow tech companies to disclaim responsibility; and since nobody else in the market is going to make any concrete moves towards a life-critical public cloud, why shouldn't they dodge whatever responsibility they can?

If my power provider fails to fix the power within a mandated period of time, tries to manipulate costs or otherwise acts in a manner counter to the good of society as a whole, my government will come down on them like the fist of an angry deity. There is very little in this world that allows for quick and easy political capital like defenestrating the power company in front of an angry populace.

Americans are not going to hold public cloud providers to account. It would take 20 years to even build the political will to create a regulatory framework wherein the public cloud providers actually had some form of meaningful legal obligation or duty of care.

There is also the part where it is in the national interests of the U.S. for the rest of the world to put all their data on servers owned and operated by corporations headquartered in the U.S. If everyone used the public cloud then the entire world would be self-delivering their data for analysis by the spooks, and the U.S. government is not going to do anything to make that harder. You don't spank the goose that's building you a golden-egg-creation machine.

What a Tool
The public cloud is not "inevitable." It will not be used for "everything." The public cloud is just one tool among many available to IT practitioners, and all but the truest of true believers will recognize this.

I don't have a backup generator for my tea kettle or my television. This is because, at the end of the day, they don't really matter. They're creature comforts that make life easier. They aren't vital to my survival.

Similarly, there are a great many workloads that aren't vital to the operation of an organization, and thus great candidates for tossing into the public cloud. What I won't put into the public cloud are the workloads that directly attach to my manufacturing machinery, or which control and monitor physical access to my premises.

If I put my accounting data in the public cloud, I'm going to make absolutely sure it's mirrored on-premises, or to another public cloud provider at a bare minimum. The same level of paranoia that was put into ensuring backups and disaster recovery capabilities for on-premises workloads must be put into ensuring backups and disaster recovery capabilities for cloud solutions.

Remove the assumption that the public cloud is always available, always on, always reliable and the economics of it change dramatically. And that, right there, is why on-premises IT isn't ever going away.

About the Author

Trevor Pott is a full-time nerd from Edmonton, Alberta, Canada. He splits his time between systems administration, technology writing, and consulting. As a consultant he helps Silicon Valley startups better understand systems administrators and how to sell to them.


Subscribe on YouTube