Amazon Web Services is always changing. New features are added and new services are launched all the time. But during re:Invent – the annual AWS conference – a flood of news is poured out on us. Some news will change the way we architect cloud-native applications today or in the coming months. While other news is irrelevant to us.
In this blog post, I highlight announcements from re:Invent 2020 with a high impact on cloud architects’ works. Let’s get started.
The following new instance types have been announced:
- mac1.metal: Yes, you can now run macOS on AWS. But only as a dedicated host at the moment.
- c6gn: AWS Graviton 2 (ARM) processor; with up to 100 Gbps networking
- g4ad: AMD processor; Radeon Pro V520 GPUs
- m5zn: Intel processor; up to 4.5 GHz; with up to 100 Gbps networking
- d3/d3en: Intel processor; up to 336 TB of local HDD storage; up to 75 Gbps of network bandwidth; up to 6.2 GiBps of disk throughput
- r5b: up to 3.6TB of NMVe instance storage; up to 60Gbps of EBS bandwidth and 260,000 IOPS
- not known: Intel / Habana Gaudi processor; optimized for training deep learning models; coming in 2021
- not known: AWS Trainium processor; optimized for deep learning training workloads; coming in 2021
A new EBS volume type was announced: gp3 (General Purpose). Volumes of type gp3 are cheaper and somehow a mixture of gp2 and io2 volumes. A gp3 volume comes with a baseline throughput, but you can also provision throughput independently of the storage size. You can start to use gp3 volumes today.
An exciting option will show up next year when io2 Block Express volumes become available. Think of them as SAN with up to 256,000 IOPS or 4,000MB/s throughput. AWS also announced multi-attach without the downside of managing concurrent writes on our own. In the end, you will be able to attach the volume to multiple instances without any headaches. We still don’t know if this works cross AZ or not.
ECS and EKS will allow you to run container workloads outside of AWS without losing the benefits of a managed AWS service: ECS Anywhere and EKS Anywhere. The biggest differences between the two options are:
ECS Anywhere runs the control plane on AWS. Therefore, you need internet connectivity to manage containers.
EKS Anywhere is powered by a Kubernetes distribution (EKS Distro) maintained by AWS. It supports deployment options with always-on, partial-on, always-off Internet connectivity.
AWS also announced managed services for Prometheus and Grafana. Grafana is a nice tool to integrate various data sources into a single dashboard (sources can be CloudWatch, Elasticsearch, Prometheus, Graphite, InfluxDB). Prometheus is a service that stores your high cardinality data (think of it as CloudWatch metrics). Prometheus seems to be way cheaper than CloudWatch in situations where a modest amount of data points are pushed per metric.
AWS Lambda now bills in 1ms increments, which will lower most bills. You now can also run container images on Lambda. Keep in mind that Lambda does not offer a managed experience that you might know from Google Cloud (aka. Cloud Run). A container that runs on Lambda has to implement the Lambda Runtime API. The container polls the Lambda Runtime API for new events to work on.
AWS Outposts will be available in smaller sizes in 2021. Instead of a whole rack, you can get an outpost with a 1U or 2U size. This will allow us to run EC2, ECS, EKS, RDS workloads in every office or store worldwide. All controlled by AWS.
CloudShell is an easy way to spin up an AWS CLI right from your browser. Click on the CloudShell button on the top right of the AWS Management Console, and a shell is started that uses the same credentials as your logged-in “user”. Works with SSO! We believe that this feature will make many enterprise users’ lives easier who sit behind corporate proxies or are not allowed to install tools such as the AWS CLI on their machines.
AWS Fault Injection Simulator is announced for 2021. Think of it as a managed ChaosMonkey. Not much is known yet about the service and its capabilities. We hope to see support for introducing network faults, EC2 faults, and much more. This will help us to run controlled experiments to see how our architected systems behave in situations that happen only rarely.
VPC Reachability Analyzer is your new tool to find an answer to the question: Why can’t EC2 instance A talk to instance B. It supports all the great features of VPCs, such as Transit Gateways and Route Tables, to figure out the problem. No more waiting for VPC flows logs to arrive.
Aurora Serverless is the future of relational databases on AWS (MySQL or PostgreSQL compatible). Unfortunately, Aurora Serverless under-delivered on the promised benefits. That’s why AWS announced Aurora Serverless v2. The new version will scale much faster, support Multi-AZ mode, read replicas, and global databases.
AWS Glue Elastic Views is a new capability that allows you to create materialized views. The source database can be a different technology than the target database. For instance, you can pull data from Elasticsearch and persist a materialized view of the data in DynamoDB for fast lookups.
Redshift will receive a bunch of machine learning (ML) features in 2021. For example, start ML predictions from your SQL query.
SageMaker received a ton of updates:
- Data Wrangler: Helps you to prepare data for machine learning
- Feature Store: Helps you to convert raw data into features for training an ML algorithm
- Pipelines: Automate all steps from data preparation up to model deployment to production
- Clarify: Detects bias in your training data and models.
- Edge Manager: Monitors your models on a fleet of devices.
- Distributed training is now supported.
We see a lot of stuff going on in the IoT space as well. Besides the usual building blocks, AWS started to announce end-to-end solutions that just work without building:
- Amazon Monitron: Order sensors and gateways from AWS. Data is automatically pushed to the AWS cloud, and you receive events if the data looks abnormal. All managed by AWS.
- Amazon Lookout for Equipment: You set up your sensors and push the data to AWS. AWS looks for abnormal values.
- Amazon Lookout for Metrics: General purpose anomaly detection in time series data.
- AWS Panorama Appliance: Connect your cameras to Panorama. The device will analyze data locally and tells you what it sees.
- AWS IoT Device Defender ML Detect: Ensure that all your devices behave as expected
AWS is getting better every year. The keynote provided a sneak peek into a ton of new capabilities that we can expect next year. It’s always good to know what’s coming if you make plans for larger projects. I hope that one of the features solves many of the troubles you expected in your next projects. Luckily, some of the features are already available and wait to be explored. Let us know how it goes!