It’s all about moving fast, and doing it now. This remains one of the key concepts of AWS and it will stay that way throughout 2020. Don’t be afraid to change up your systems as you will fall behind quicker than you realize. Like last year, Andy Jassy’s keynote was full of new releases and a bunch of developers getting excited about new ways to put things in databases. Closing out 2019, AWS remains te largest vendor with close to 45% market share. Microsoft is in the rearview mirror, but far far away.
AWS released a new generation of ARM-based processors, which deliver a 40% better price per performance compared to the latest GPU based instances. Andy also used a lot of words that I did not understand, so to learn more about AWS Nitro, you can start here.
84% of the Kubernetes workloads run on AWS. This is a massive amount, almost as big as the amount of developers that don’t care about servers. To solve this problem, AWS launched Fargate last year. Fargate solves the problem of provisioning servers to run containerized workloads by provioning servers for you under the hood.
With the release of Fargate support for EKS, you can now run your Kubernetes workloads on Fargate. So you can stop worrying about servers and clusters, while you can still question every choice you’ve ever made including the one about using Kubernetes.
- AWS Transit Gateway Multicast NEW
- Accelerated site to site VPN NEW
- Transit gateway inter-region peering NEW
- Transit gateway network manager NEW
Moving to the cloud is like moving house. What are the things you take with you and what are the things you leave behind ? - Andy Jassy, 2019
On-premises cost a lot of money for the things you get. You have to spend a lot of money without getting the same features and security that the cloud can bring. Every industry is moving away from mainframes. They’re slow, complicated and expensive. No one has mainframe knowledge anymore. Some companies pick the whole thing apart and launch them as microservices on AWS, some pick the things they need and move them to AWS. People are trying to move away from Oracle and SQL server; They’re expensive. Microsoft is not prioritising what matters to the customers, look at all the damn licences. People are moving to open engines like MySQL, MariaDB and Postgres. AWS offers same performance with open engines using AuroraDB. It is the fastest growing service on AWS.
People are moving from Windows to Linux. People don’t want to pay the price anymore for Windows. The Linux community is a lot more sophisticated. People are not very happy when one entity owns the operating system. Looking at you, Bill Gates.
Not your father’s data requirements. - Andy Jassy, 2019
Companies are moving from data silos to datalakes to make machine learning and analytics much easier. Data lakes are built on top of S3, because it is more reliable and scalable than anything else. Lots of this has to do with multi-AZ magic. Other vendors offer low-grade availability zones because they’re right next to eachother in the same building. S3 is the only object store to block public access on multiple levels and get an inventory on objects. It’s the easiest object store to get data into through a million options.
Read more about S3 intelligent tiering and S3 batch operations. S3 deep archive costs less than a dollar per TB, which makes it the perfect candidate to store all your backups and other data that may take a while to retrieve, but must be stored for a long time.
So you finally figured out how to implement security and access control on your S3 buckets? Your bucket policy is a work of art and your consumer is working flawlessly? But then you need to add a second client, and a third, and before you know, your bucket policy is a mile long (or 1,60934 kilometers for users of the proper metric system). Introducing S3 access points, AWS claims to simplify access control to your bucket over a variety of clients.
Read more about S3 access points
Not your father’s database. - Andy Jassy, 2019
Disk will fill in 12 hours. Your pager beeps, your heart rate rises, this alert sends shivers down your spine. You’ve been trained for this, you know what to do, so you turn on your laptop and open the AWS console. You navigate to the Redshift console and look for a way to increase disk space. And then it hits you: you have to increase the instance type aswell. This doesn’t have to be reality and luckily AWS had the same thoughts. With the new RA3 instances, you can scale storage and compute independently.
It’s hard to scale without compromising performance. SSD throughput is scaling at a way faster rate than CPU processing rate.
Hardware accelerated cache. Move the compute to the storage. This delivers 10x better query performance than anywhere else.
Please let it be Cloudwatch Kibana 🤞 - Kenneth De Win, 2019
Scale up to 3 PB of log data per cluster. Move unused data to S3. Save 90% of costs compared to what you do in ES today. If a query needs to pull data from S3, some Nitro magic make the data accessible very quickly.
😭 - Kenneth De Win, 2019
Please be managed Casandra 🤞 - Rob Kenis, 2019
No clusters to be managed, pay on demand, uses same Cassandra drivers. There’s not much more to it. Finally use Cassandra without contemplating your career as a developer.
🥳 - Rob Kenis, 2019
Machine learning continues to grow on AWS. All types of developers like to try it sooner or later, this resulted in 3 layers of services.
AWS optimized TensorFlow is 20% faster than any other option (some very fancy private hardware not open to the public). Optimized PyTorch is 22% faster, just like MXNet.
Fully integrated development enviroment for SageMaker magic. Web-based IDE. This makes it much easier to write models. Read more
One-click notebooks with Elatic Compute. No instances to provision, increase or decrease compute resources without interruption.
Creating and training a new model requires a lot of little tweaks and fixes. SageMaker Experiments makes this process fun again.
Real-time metrics of model performance, help understand and interpret models. Results can be viewed in SageMaker Studio.
Looking at a model is like looking at a compiled binary. It makes no sense to the naked eye. - Andy Jassy, 2019
Your model might have been pretty accurate last year, but something changed and now your results are far from reality. Detect concept drift in deployed models with SageMaker Model Monitory. Analyze and visualize models so you can detect concept drift.
tl;dr: data -> 🧙 -> model.
Customers want automatic models with more insight in the actual model.
Provide a csv with training data, Autopilot trains 50 unique models. SageMaker provides all the notebooks so you can tell how it actually works. All algorithms are ranked in a model leaderboard so you can choose which one you want. Autopilot gives you full control to tweak the models.
SageMaker studio is a giant leap forward being the first machine learning IDE - Someone at AWS
Collect all data and put a csv in S3. Create a new autopilot project in SageMaker Studio with the csv. SageMaker starts training 50 models and gives you an overview so you can pick the best one. You can look at the exact algorithms used by Autopilot with the Debugger information provided by SageMaker Debugger. After you pick your favorite model, you can deploy it right from SageMaker Studio. If SageMaker monitor finds any drift, it gives you an alert so you can train another set of models with new, fresh data.
Detecting fraud is very expensive and doesn’t scale well because it uses little to no ML.
Send AWS your transactions and purchase history, they build a unique model for you. That model will detect certain email domains for example based on the Amazon experience in fraud detection. These models will give you a fraud scoring on certain events.
Amazon CodeGuru is a machine learning service for automated code reviews and application performance recommendations. It helps you find the most expensive lines of code that hurt application performance and keep you up all night troubleshooting, then gives you specific recommendations to fix or improve your code. You can “call” CodeGuru by adding it as a reviewer on your Pull Request (currently only CodeCommit and GitHub). It then checks your code based on learning models that are trained on Amazon’s code bases comprising hundreds of thousands of internal projects, as well as over 10,000 open source projects in GitHub.
CodeGuru only supports Java applications today, with support for more languages coming soon.
Generates analytics on Amazon Connect, which is a managed contact center. Contact Lens captures calls and analyzes if the conversation was positive or negative. With enough data, you can create dashboards containing overall satisfaction.
Amazon Kendra is a highly accurate and easy to use enterprise search service that’s powered by machine learning. Kendra delivers search capabilities to your website or application. Kendra lets you easily add content from file systems or external data sources like Jira or Google Drive. You can also provide a set of FAQs to make it even easier for your users. After uploading the data, Kendra syncs and indexes the data. It’ll find concepts and relations between content. The console offers the ability to test and refine queries. Natural language enables you to get more specific answers from anywhere in your data. Ask questions like, “Is the IT help desk open at noon?” or “How do I connect to my VPN?” so you can get more precise answers, instead of some terrible confluence page.
AWS Outposts is a fully managed service that extends AWS infrastructure, AWS services, APIs and tools to virtually any datacenter or on-premises facility for a truly consistent hybrid experience. AWS Outposts is ideal for workloads that require low latency access to on-premises systems, local data processing, or local data storage.
With AWS Outposts you can run Amazon EC2, Amazon EBS or Amazon RDS in your on-premise infrastructue. Amazon S3 will be available in 2020.
A Local Zone is a new type of AWS infrastructure deployment that brings a couple of AWS services very close to a particular geographic area. It’s basically a smaller and more local version of an AWS region. They’re designed to provide very low (single-digit milliseconds) latency to applications. Right now, they’re only available in Los Angeles, but AWS will expand to other locations in 2020.
AWS Wavelength enables you to build applications that serve mobile devices with single-digit millisecond latencies over 5G networks. These applications can include games, live video streaming or virtual reality. It brings AWS services to the edge of the 5G network, minimizing the network hops and latency to connect to an application from a 5G device.
A lot of people get nervous about change. It’s scary how far behind you can become in a very short period of time. With what the cloud offers, you can build things that were never possible before. That is an opportunity unlike any other.