[I’m writing a longer piece about cloud computing from a personal and political economy perspective – because I was there at the start and remain in the thick of it as a ‘cloud architect’ so I have both thoughts and experience…in this short piece, I use the recent Amazon Web Services disruption to discuss the role of power in ‘cloud’ or what should more precisely be called utility computing]


The 7 Dec 2021 Amazon Web Services (or, AWS) ‘outage’ has brought the use of cloud computing generally, and the role of Amazon in the cloud computing market specifically, to the attention of a general, non-technical audience [btw, outage is in single quotes to appease the techies who’ll shout: it’s a global platform, it didn’t go down, there was a regional issue! and so on]

Outage, in the total sense, or not, the event impacted a large number of companies, many of which are global content providers such as Disney and Netflix, services such as Ring and even Amazon’s internal processes that utilize their computational infrastructure.

Before the cloud era, each of these companies might have made large investments in maintaining their own data centers to host the computers, storage and networking equipment required to host a Disney+ or HBOMAX platform. In the second decade of the 2000s (really gaining momentum around 2016) the use of at first, Amazon Web Services and then Microsoft’s Azure and Google’s Cloud Platform offered companies the ability to reduce – or even eliminate – the need to support a large technological infrastructure to fulfill the command and control functions computation provides for capitalist enterprises.

Computation, storage and database – the three building blocks of all complex platforms – are now available as a utility, consumable in a way, not entirely different from the consumption of electricity or water (an imperfect analogy since, depending on the type of cloud service used, more or less technical effort is required to tailor the utility portfolio to an organization’s needs).


What is Cloud Computing? What is it’s Political Economy? What are the Power Dynamics?

Popular Critical Meme from Earlier in the Cloud Era

A full consideration of the technical aspects of cloud computing would make this piece go from short(ish) to a full position paper (something I’ll address in that bigger essay I’m working on which I mentioned at the top). So, let’s answer the ‘what’ question by referring to what’s considered the urtext within the industry: the NIST definition of cloud computing

Cloud computing is a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction. This cloud model is composed of five essential characteristics, three service models, and four deployment models.

https://csrc.nist.gov/publications/detail/sp/800-145/final

The NIST document goes on to define the foundational service types and behaviors:

  • SaaSSoftware as a Service (think Microsoft 365 or any of the other web-based, subscription services that stop working if your credit card is rejected)
  • PaaSPlatform as a Service (popular industry examples are databases such as Amazon’s DynamoDB, Azure SQL or Google Cloud SQL)
  • IaaSInfrastructure as a Service (commonly used to create what are called virtual machines – servers – on a cloud platform instead of within a system hosted by a company in their own data center)
  • On-demand Self Service (which means, instead of having to get on the phone to Amazon saying, ‘hey, can you create a database for me’ you can do it yourself using the tools available on the platform
  • Reserve Pooling – (basically, there are always resources available for you to use – this is a big deal because running out of available resources is a common problem for companies that roll their own systems)
  • Rapid Elasticity – (have you ever connected to a website, maybe for a bank and have it slow to a crawl or become unresponsive? That system is probably stressed by demand beyond its ability to respond. Elasticity is designed to solve this problem and it’s one of the key advantages of cloud platforms)
  • Measured Service – (usage determines cost which is a new development in information technology. Finance geeks – and moi! – call this OPEX or operational expense and you better believe that beyond providing a link I’m not getting into that now)

To provide a nice picture which I’m happy to describe in detail if you want (hit me up on Twitter) here’s what a cloud architecture looks like (from the AWS reference architecture library):

AWS Content Analysis Reference Architecture

There are a lot of icons and technical terms in that visual which we don’t need to get into now (if you’re curious, here’s a link to the service catalog). The main takeaway is that with a cloud platform – in this case AWS but this is equally true of its competitors – it’s possible to assemble service elements into an architecture that performs a function (or many functions). Before the cloud era, this would have required ordering servers, installing them in data centers, keeping those systems cool and various other maintenance tasks that still occasionally give me nightmares from my glorious past.

Check out this picture of a data center from Wikipedia. I know these spaces very well indeed:

Data Center (from Wikipedia)

And to be clear, just because these reference architectures exist (and can be deployed – or, installed ) that does not mean an organization is restricted to specific designs. There’s a toolbox from which you can pull what you need, designing custom solutions.

So, perhaps now you can understand why Disney, for example, when deciding to build a content delivery platform, chose to create it using a cloud platform – which enables rapid deployment and elastic response instead of creating their own infrastructure which they’d have to manage.

Of course, this comes with a price (and I’m not just talking about cash money).

Computer Power is Power and the Concentration of that Power is Hyper Power

Now we get to the meat of the argument which I’ll bullet point for clarity:

  • Computer power is power (indeed, it is one of the critical command and control elements of modern capitalist activity)
  • The concentration of computer power into fewer hands has both operational and political consequences (the operational consequences were on display during the 8 December AWS outage (yeah, I’m calling it an outage cloud partisans, deal)
  • The political consequences of the concentration of computer power is the creation of critical infrastructure in private hands – a super structure of technical capability that surrounds the power of other elements of capitalist relationships.

To illustrate what I mean, consider this simple diagram which shows how computer capacity has traditionally been distributed:

Note how every company, with its own data center, is a self-contained world of computing power. The cloud era introduces this situation:

Note the common dependency on a service provider. The cloud savvy in the audience will now shout, in near unison: ‘but if organizations follow good architectural principles and distribute their workloads across regions within the same cloud provider for resiliency and fault tolerance (yes, we talk this way) there wouldn’t be an outage!’

What they’re referring to is this:

AWS Global Infrastructure Map Showing (approximate) Data Center Locations

From a purely technical perspective, the possibility of minimizing (or perhaps even avoiding) service disruption by designing an application – for example, a streaming service – to come from a variety of infrastructural locations, while true, entirely misses the point…

Which is that the cloud era represents the shift of a key element of power from a broadly distributed collection of organizations to, increasingly, a few North American cloud providers.

This has broader implications which I’ll explore in greater detail in my upcoming piece.

UPDATE 11 Dec

Amazon has posted an explanation (which, in the industry is known as a root cause analysis) explaining the outage. I’ll be digging into this in detail soon.