At this year’s re:Invent, the big theme from AWS was that its never-ending cadence of innovation gives developers (aka builders) the right tools to get the job done. There was great focus on machine learning, databases, hybrid cloud, and account management.
Compute-related: AWS released a pair of new virtual machine instance types. The EC2 A1 instance feature ARM-based processors and are suitable for Linux workloads. If you need virtual machines with up to 100Gbps of network bandwidth, the new C5n instances are now available. EC2 users who want to scale quickly without warming up new VMs can now hibernate a running (Amazon Linux) instance and wake it up when needed. Users only pay for storage and Elastic IPs when a VM is in a hibernated state. AWS also shared their intent to build a hybrid computing experience. Not available until later in 2019, AWS Outposts are intended to bring AWS services and infrastructure to any data center.
Serverless computing got a boost at re:Invent. AWS announced Lambda Layers which gives developers up to five packages to inject into a given function at run time. This promotes component sharing and keeps the core function package as small as possible. Lambda users also get access to custom runtimes that opens up the option to use languages, such as C++ and Rust for functions. For Lambda functions serving up HTTP(S) requests, you can now use the Application Load Balancer. The function-oriented workflow service, AWS Step Functions, also got a refresh with new integrations to services like DynamoDB, AWS Batch, Amazon SQS, and Amazon SageMaker. Finally, AWS announced Firecracker, an open source virtualization technology based on KVM that’s used by Lambda for sandboxing functions.
For networking, AWS introduced a global traffic router called AWS Global Accelerator. This service uses static Anycast IP addresses as an entry point and routes TCP or UDP requests through the AWS global network based on characteristics such as geography, application health, and custom weighting rules. The new AWS Transit Gateway service makes it possible to use a single gateway for connecting Virtual Private Clouds (VPCs) to on-premises networks. This gets customers away from point-to-point connectivity and into a hub-and-spoke architecture. Service meshes are a hot topic nowadays, and AWS App Mesh is Amazon’s answer. This beta service uses the Envoy proxy (and an AWS control plane instead of Istio) to improve control flow and observability of microservices. As a related service, AWS Cloud Map provides service discovery by tracking application components and helping developers discover services at runtime.
Storage: S3 is the AWS object storage service and it received a few key updates. First, it got Intelligent Tiering. Today, S3 users choose among storage classes based on how frequently they need to access their data. With Intelligent Tiering, users have a new storage class that monitors access patterns and moves infrequently-accessed objects to a cheaper tier. S3 also now has a preview feature (S3 Batch Operations) that simplifies bulk operations against objects in S3. AWS also added the ability to “lock” an object during a customer-defined period to support data retention requirements. For those working with the file system versus object storage, AWS offered a pair of relevant announcements.
Amazon FSx for Lustre offers a managed, distributed file system for compute-intensive workloads. And, Amazon FSx for Windows File Server delivers a managed Windows file system for workloads that need file storage. It supports the SMB protocol, Windows NTFS and Active Directory integration. For those looking to move chunks of data from on premises to Amazon S3 or Elastic File System, the new AWS DataSync is available now.
Database and Analytics-related: AWS offered a pair of key updates to their NoSQL database, DynamoDB. First, DynamoDB Transactions offer ACID-compliant transactions across multiple tables in a given AWS region. Secondly, DynamoDB On-demand is a new pricing model that doesn’t require up-front capacity planning, while still offering all the standard DynamoDB features. Their relational database, Amazon Aurora, got a new geo-replication feature called Amazon Aurora Global Database. This applies to the MySQL variant of Aurora, and creates read replicas that can quickly be promoted to a primary in the case of a regional outage. AWS also announced an upcoming preview of Amazon Timestream, a time-series database with no infrastructure management requirements.
Regarding data analytics, AWS Lake Formation (not yet in preview) is for creating S3-based data lakes from a variety of data sources. Another preview service is ML Insights for Amazon QuickSight which offers things like anomaly detection and forecasting. For data processing, AWS announced Amazon Managed Streaming for Kafka (in preview) where developers can get managed Kafka clusters (including Zookeeper clusters).
Blockchain: Amazon announced the new Amazon Quantum Ledger Database (QLDB) which is a managed ledger database that offers an immutable log. It complements the upcoming preview of Amazon Managed Blockchain where the blockchain network activity can be replicated into QLDB.
Apps: For container-based deployments, users can now leverage AWS CodeDeploy for blue/green deployments to minimize downtime. The AWS Transfer for SFTP offers a highly available managed service for getting data into Amazon S3 buckets. The set of people excited about managed SFTP and those excited about robotics probably has little overlap, but the latter group now has AWS RoboMaker for developing, simulating, and deploying robotic apps at scale. Finally, Amazon Personalize (in preview) uses machine learning to give developers a capability to build and consume recommendation models.
Internet-of-Things: AWS IoT Events is a new managed service for detecting and responding to events from IoT sensors and apps. It detects events across thousands of sensor types. The AWS IoT Things Graph lets developers visually connect devices and services to create IoT apps. It includes prebuilt models for popular device types, or you can build your own custom model. And AWS IoT SiteWise is designed to work with industrial equipment. Your on-premises industrial data goes through a gateway and is stored in AWS for analysis.
ML/AI: Amazon announced that their own “Machine Learning University” would be available to any developer. Then, AWS kicked off a series of announcements related to AI/ML services, particularly around Amazon SageMaker. The new SageMaker Ground Truth capability helps you label datasets used to train machine learning systems. Options for labeling include automation via active learning, or Amazon Mechanical Turk for human intervention. SageMaker Neo is a new feature that makes it possible to train machine learning models once, and run them anywhere in the cloud or edge. This supports a variety of frameworks (e.g., TensorFlow, PyTorch) and hardware architectures (e.g., ARM, Intel). For data scientists with limited data sets for training and an interest in reinforced learning, there’s now SageMaker RL. And you can also now use Git repos to store SageMaker notebooks. Additional ML announcements included the preview of Amazon Textractfor optical character recognition (OCR) and Amazon Elastic Inference which lets you add GPU acceleration to any EC2 (and SageMaker) instance. Finally, AWS announced a new “machine learning” category of products (representing algorithm and model packages) in the AWS marketplace.
Management: AWS Control Tower (in preview) makes it easier to set up multi-account AWS environments with blueprints to deploy components in a secure fashion. The AWS Security Hub, also in Preview, gives you a centralized view of security alerts across accounts, as well as constant checks of your service configurations. If you struggle with managing all the commercial software licenses used in your account, the AWS License Manager is designed to help. And if companies want to limit the catalogue of add-on services available to users, the AWS Private Marketplace should add value. Finally, the AWS Well-Architected Tool offers a self-service experience for reviewing AWS workloads to see if they match best practices.