I recently helped out with the inaugural Serverless Conf, and gave the closing remarks. The event was held in a warehouse in Williamsburg, New York. Organiser Ant Stanley of ACloudGuru did a great job of organising his first conference.
There are plenty of smarter people than me explaining what “serverless” is – Paul Johnston, who spoke at the conference, has a great blog series on the topic. To me however serverless computing is primarily interesting for three key reasons:
- On demand computing is taken to its natural conclusion – paying at the point of customer value. AWS Lambda, the first mainstream serverless platform, runs compute services in response to events – you write and upload code to AWS Lamda, associating it with particular AWS services, but what is kind of mind-blowing is that you don’t pay for the services until they are actually invoked. A startup can build out an entire infrastructure, without paying anything for it until the customer does.
- The atomic unit of compute is the function. Just as we begin to understand the implications of container-based computing, in which the atomic unit of compute is the application, comes a new model to consider, in which the individual function is the service. As with microservices this kind of approach creates coordination overheads, but it also offers a great deal of flexibility.
- You don’t need to think about running containers, any more than you do running an operating system when developing to a PaaS. Unlike containers, which have been associated closely with microservices, serverless is more like PaaS in abstracting all infrastructure required to run a service. Serverless certainly doesn’t obviate the need for ops, but it does exemplify a more PaaS like model.
The chart above suggests a couple of interesting things about the discontinuity in 2007. Deal size shrunk, while average number of deals dramatically increased. Why? The obvious answer is the launch of AWS, and the attendant creation of Y Combinator and the accelerator model, rather than traditional VC-driven company creation. Given that serverless hasn’t exploded so far, it might seem overblown to draw comparisons with the introduction of AWS. And yet. And yet. There is no doubt that today smaller teams can make a bigger impact than ever before. Consider Instagram or WhatApp, which sold to Facebook for $1.9bn with only 55 employees. Serverless attempts to industrialise developer impact- if we’re going to see the first single employee billion user multi-billion dollar valuation startup it’s likely they’ll be building on something like AWS Lambda. Could we see the another similar discontinuity to the graph above, with deal size cratering again, and many more flowers blooming? It’s possibly. Certainly Amazon’s serverless implementation would hope to enable that.
As Sam Kroonenberg, CTO of ACloudGuru put it:
“AWS Lambda is to compute what s3 is to storage.”
AWS is pretty confident about serverless advantages. It’s Serverless GM Tim Wagner made it pretty clear that he expects serverless to put severe pressure on orthogonal markets – such as Docker-based infrastructure, partly because he expects the model to drive more testing into the cloud. Serverless is not solely an Amazon phenomenon however. IBM has adopted an open source model with OpenWhisk, while Google is building a set of services around Firebase, originally a database as a service it acquired. Microsoft has taken a hybrid hosted/open source approach with Azure functions.
Competition has been sufficient that Amazon took the surprising step of announcing it was to open source a serverless framework called Flourish. On that more details as they emerge.
Joe Emison works at DMGT building B2B applications and services. He gave a great presentation which was frankly pretty scathing about AWS.
“In this case i don’t think Amazon has the best option for any of the services out there. It’s about the front end. AWS serverless is largely about back end processing, which we have largely outsourced. To me, the advantage of serverless is the key is faster iterations, and a move to fatter clients”
30 day trials of SaaS products that expire before you’ve had a chance to schedule some time to actually try them. Such an anti-pattern.
— Stef Lewandowski (@stef) July 1, 2016
(Read this and other great posts @ RedMonk)