• United States

AWS re:Invent and the 5 fronts of the cloud arms race

Dec 07, 20176 mins
Artificial IntelligenceCloud ComputingContainers

Increasingly, the public cloud arms race is being waged on four fronts, with a fifth quickly emerging. All five had a healthy set of announcements at this year's event.

AWS re:Invent
Credit: Amazon

For the last six years running, the most important event in cloud computing has been AWS re:Invent, where the market leader announces its latest improvements. This year, 44,000 people descended upon a very crowded set of Las Vegas venues spread across multiple hotels for breakout sessions, certification exams, a diverse expo floor, and the all-important keynotes where the newest offerings were announced.

Increasingly, the public cloud arms race is being waged on four fronts, with a fifth quickly emerging. All five had a healthy set of announcements—here are some of the highlights.

1. IaaS/PaaS

AWS started the cloud revolution with its S3 object storage service in 2006, which was quickly followed by its EC2 compute offering and a set of other IaaS products. As time went by, PaaS services like load balancers, message queues, and databases emerged as key components as well. Both classifications of services are, of course, built on physical hardware that AWS organizes into availability zones and regions.

That physical footprint, AWS announced, will grow to span 22 regions and 107 point-of-presence locations for its edge-caching network by the end of 2018. AWS likes to point out that their definition of an availability zone spans multiple data centers, each of which is separated by a “meaningful difference,” which often is more elaborate than competitive offerings that use similar terms. The impressive part about this is that in its first 10 years, AWS made 11 regions available, and in the subsequent three doubled that number, which clearly speaks to the speed at which they can rack equipment.

While their new security service in this space is equally impressive, the big announcement on this front is the introduction of bare metal instances. Over time, AWS has been able to offload more and more virtualization layers onto specialized devices they’ve built in the wake of their Annapurna acquisition, which sit alongside physical servers to the point where there is little enough overhead to make these new bare metal instances possible. That especially helps workloads that need access to things like performance counters that aren’t reachable on virtual machines.

2. AI

IBM Watson put a face to services like image recognition, sentiment analysis, and speech-to-text (and back), and AWS found itself in the unusual position of playing catch-up on offerings like this that require deep learning back ends. On this front, not only did AWS make a series of announcements that get them closer to parity on those services, but they showed off a new video recognition service that goes well beyond simpler image recognition to include person tracking, even when a person is blocked in a set of frames. Not only that, but they announced a new piece of hardware called DeepLense that developers can use to play around with the service.

But AWS didn’t start there when it comes to AI. There’s a steep learning curve to set up machine learning environments and their new SageMaker service makes it far easier to create and configure the necessary pieces. This includes collecting and preparing training data, choosing and optimizing a machine learning algorithm, setting up the training environments, training and tuning the model, deploying the model in production, and managing that production environment and scaling it when necessary. They are also offering a smaller version of this machine learning environment on GreenGrass and packaging both onto DeepLense.

3. Serverless

When AWS announced Lambda at this same conference a few years ago, most people (including me) didn’t know what to make of it. But more recently, Function-as-a-Service has become the popular keystone to the serverless architecture movement that makes use of other services like DynamoDB, API Gateway, and S3’s web hosting mode to enable applications to achieve a much finer-grained usage cost model.

To appeal to a wider audience, though, serverless has suffered from data and IDE issues that AWS addressed nicely. DynamoDB, the NoSQL solution, now offers something called Global Tables, which replicates data across multiple regions so that serverless applications can more easily achieve high availability. Similarly, some applications require traditional relational database support and with that in mind, AWS pre-announced a serverless payment model for its existing Aurora relational database.

To lower the serverless learning curve for developers, among the features of the new Cloud9 IDE is that it provides Lambda debugging. So that engineers can more easily share components with one another, AWS also launched a function repository, which is akin to a DockerHub for Lanbda.

4. Kubernetes

By far the most surprising announcements at re:Invent this year involved Kubernetes. It was surprising because AWS holds such strong leads in so many fronts, and given their heavy investment in Serverless, it wasn’t outlandish to assume that they’d concede the container world to Google and Microsoft.

Instead, AWS announced a second container offering in parallel to the existing ECS that runs Kubernetes clusters called EKS. Further, they announced Fargate, which automates the configuration and spin-up of a container cluster so that developers can focus on the code placed within it instead of standing it up and maintaining it. Together, the two services show AWS’ commitment to container-based application development even as they innovate on its successor in Serverless.

5. IoT

Of the five fronts described here, IoT is the most undefined. Meaning, everybody seems to agree that it’s an important topic, but not everybody is quite in agreement on how it’ll all play out. After a set of introductory IoT offerings last year, AWS iterated over them nicely with announcements centered around security, analytics, and scalability. All three of these aspects will be important as IoT matures, given the difficulty of managing hundreds of thousands—if not millions—of devices. How much processing of data will occur on the edge vs. the public or private cloud has yet to be determined – and may vary greatly from use case to use case – but AWS has a nice set of foundational functionality here to build off of.

And so, another year of announcements demonstrating AWS’ impressive innovation pace came to a close. While the crowds were at times unmanageable and there were some awkward moments that bordered on cultural appropriation in the Andy Jassy keynote, the Werner Vogels keynote did a nice job of putting a bright spotlight on three genuinely brilliant engineers in what can only help the gender inequality that plagues the tech industry in general. AWS Software Engineer Clare Liguori led an excellent demo of the new IDE offering, Netflix Senior Chaos Engineer (and how cool a job title is that?) Nora Jones discussed lessons learned from advanced quality measures, and AWS Senior Technical Evangelist Abby Fuller stole the show on the final day with her enthusiastic and insightful thoughts on Kubernetes topics that included a demonstration of Fargate. More, please.


Pete Johnson is a Technical Solutions Architect at Cisco, covering cloud and serverless technologies. Prior to joining Cisco as part of the CliQr acquisition, Pete worked at Hewlett-Packard for 20 years where he served as the Chief Architect and as an inaugural member of the HP Cloud team.

The opinions expressed in this blog are those of Pete Johnson and do not necessarily represent those of IDG Communications, Inc., its parent, subsidiary or affiliated companies.