When AWS introduced its Simple Storage Service and Elastic Compute Cloud in March and August of 2006, respectively, it was not immediately obvious that cloud services would come to dominate the industry. Indeed, even years after their introduction many long time enterprise technology vendors still regarded them and the other AWS services that followed as little more than toys, unsuitable for and certainly uninteresting to the enterprise.
What was obvious, however, was that whatever one thought of the cloud services themselves, the economic model that accompanied them represented a sea change. Up until that point, the dominant model for enterprises was purchased physical hardware. And while many providers from 1&1 to Rackspace offered hosted machines, these were typically rented by the month and often included setup or instantiation charges to discourage trivial experimentation.
With the introduction of EC2 and S3, it became possible to spin up a machine in minutes, run it for an hour, and then kill it all via the basic combination of a browser and a credit card – and only pay for the hour. As Flip Kromer put it back in 2009, “EC2 means anyone with a $10 bill can rent a 10-machine cluster with 1TB of distributed storage for 8 hours.”
For those who have grown up in an era characterized by the ready and expanding availability of cloud resources, it’s difficult to convey precisely how game changing this model was. The current market valuations of Amazon, Microsoft and others imply it, but the cloud’s pay-as-you-go economic model allowed developers that already had a large library of open source software available at no cost access to hardware they could afford themselves.
In a relatively short span of time, then, the technology landscape shifted from a CAPEX oriented model to one much more OPEX-centric. The market has never been the same since.
If we allow, then, that one of the more lasting impacts of the advent of cloud services is the impact to pricing models, that raises an interesting question: will the same be true of serverless?
Serverless is a term with a history that is longer than is commonly understood; it was used for years as a generic descriptor. For example, it was cited in conjunction with network filesystems in this UC Berkeley paper from 1995, virtual reality in 2004, databases (specifically SQLite) in 2005, RFID authentication in 2008 – even printing in 2012.
Whatever it had been before, however, the term serverless was effectively coopted by AWS in 2014 with the introduction of Lambda. While that service took somewhat longer to take off than originally predicted, its function-oriented model became synonymous with serverless overnight – conversations about serverless were de facto conversations about functions.
Interestingly, however, this strictly function-oriented definition for serverless has begun to blur in recent years with the consent of all parties involved. As AWS competitors, it was always in Google and Microsoft’s best interests to avoid a definition of serverless that was closely tied to Lambda. Given their respective services’ adoption deficits relative to AWS’ market leading product, allowing serverless to be a synonym for functions meant effectively ceding the term – which was exploding in popularity – to the market incumbent. Instead, the two competitors for AWS’ position sought every opportunity to expand serverless conversations well beyond purely functional architectures, because they stood a better chance of competing on those terms.
More curious has been the willingness of AWS itself to be loose with its usage of serverless. While tactically it would be natural to expect Amazon to jealously guard usage of the rebooted serverless term to advantage Lambda from a marketing standpoint, instead the company has been willing to apply it to databases and other non-function oriented services.
With all three hyperscalers converging on a broader, more inclusive scope for serverless, then, a new definition would appear to be in order. From RedMonk’s perspective, as my colleague has captured, the most accurate description of this evolving serverless category is something close “managed services that scale to zero.” Importantly, this definition implies growth not necessarily of the original functions-based architecture but rather of the economic model that accompanied it.
The cloud’s pay-as-you-go model, with apologies to predecessors such as time sharing, heralded a revolution in pricing, consumption and infrastructure and redefined the industry in its image. It had such an advantage in accessibility over the model that preceded it that not offering a pay-as-you-go option was not, itself, an option.
The real question now about serverless then is not about the technology, but whether its pay-only-when-it’s-on economic model is about to do exactly the same thing that the cloud’s did before it.
Disclosure: Amazon and Microsoft are RedMonk clients. Google is not currently a client.
