Re:Invent, AWS’ annual tech conference in Las Vegas, was attended by 43,000+ people this year and I was there to share ideas and learn about the latest developments in the platform.
The stand out theme from this year’s re:Invent was machine learning and how to make inferences from datasets. Many of the customer testimonials made reference to machine learning, such as the NFL’s use of ML in their Next Gen Stats program or Digital Globe’s use of AWS SageMaker to predict which satellite images a user will likely want to look at next and prefetch it from AWS Glacier for faster load times, whilst maintaining cost efficiency. Deeplens, Amazon’s $249 intelligent camera with onboard compute, aims to act as a devkit for developers to give them a platform to experiment with ML.
It was quite clear that AWS was all too aware that Google’s cloud offering had some USPs over AWS, and some of the announcements at re:Invent like EKS and inter-region VPC peering were to address the gaps in the existing feature set.
Here are the headline announcements from this year’s re:Invent:
It only takes a look at how popular the Tensorflow Github repository is to see that machine learning is widely used in many companies, and the big announcement by AWS in this field was Amazon SageMaker. Amazon SageMaker is a managed service to manage the building, training and deployment of machine learning models and, like many of the AWS’ most popular services, it’s designed to reduce friction for ML developers and data scientists.
Aurora, AWS’ MySQL and PostgreSQL compatible relational database service, received some notable new features. Aurora has always allowed for spreading reads across up to 15 replicas easily. Aurora Multi-Master allows for the same horizontal scaling in a single region for writes as well, with multi-region writes coming in 2018. Aurora is a common building block in our customer deployments and Multi-Master is a feature that will make designing highly scalable database tiers less complicated to deploy and manage.
Aurora Serverless takes advantage of Aurora’s separation of the storage and processing components of the architecture to provide rapidly scaling on demand compute. Aurora Serverless pricing is based on capacity units similar to the model employed by DynamoDB, which lends itself to spiky workloads.
In the NoSQL space, AWS announced DynamoDB Global Tables which automatically replicates a DynamoDB table into multi-regions for high availability and performance improvements for globally distributed systems.
AWS also extended its managed database offering by launching Amazon Neptune, a managed graph database for datasets with interconnected relationships.
AWS Lambda added support for Go and doubled the memory limit. Although these aren’t necessarily the most headline-grabbing announcements, Lambda featured in most of the customer case studies and discussions—it’s clear that the serverless revolution is powering ahead at full steam, and its adoption is only accelerating.
Kubernetes has grown massively in the last year, so it wasn’t surprising to see AWS announce a managed service for this. AWS EKS is a managed service for deploying a highly available Kubernetes cluster with added integration with AWS IAM and VPC. Kubernetes is probably the #1 requested technology from our customers, and we manage many Kubernetes clusters on AWS, so EKS is an announcement we’re very excited about. We’re looking forward to the closer integrations with IAM and VPC.
AWS Fargate was announced which abstracts the user away from the container platform entirely, so a container can be launched without having to worry about the host running the container. Fargate means developers can focus on developing their application and leave the hosting and scaling of their application container to AWS.
During his keynote, AWS CEO Andy Jassy said:
"Of all the buzzwords of choice...that we've been working on at AWS, IoT might be delivering the fastest in terms of the actual number of companies doing real work there."
In acknowledgement of this growth, AWS announced several new IoT services. AWS IoT 1-Click makes it easier for simple connected buttons to trigger Lambda functions. AWS IoT Device Management, AWS IoT Analytics and AWS IoT Device Defender are geared towards making the process of running and maintaining a fleet of IoT devices easier. Finally, Amazon FreeRTOS was announced which is an extension to an operating system for microcontrollers to help add cloud connectivity to low-cost IoT devices built on popular chipsets.
Inter-region VPC Peering
This was only mentioned in passing by Werner Vogels in his keynote but I’m excited enough about it to call it out as a highlight. VPCs in separate regions can now be peered together, allowing for multi-region network comms over AWS’ network backbone. Previously, to communicate between VPCs in different regions, complicated VPN tunnels would need to be set up and communication would travel over the internet. This announcement means multi-region deployments are much less complex and more robust.