If you are reading this you already know 2 things:
- You know what re:invent is.
- You know it is virtually impossible to sum up the reinvent announcements, let alone as it is taking place.
No argument there. Still, what I was able to do is gather a little over 20 announcements I found interesting and which I think everyone in the tech, startup and VC industry should be aware of.
I gathered them into 7 topics: Compute, Storage, Deep Learning, Serverless, IoT, Hybrid/Multi Cloud Architecture, AI Verticals.
You’ll find links under each of these, leading to additional content, provided either by AWS or another resource that covered each of the announcements well, in my humble opinion
I am also aware many of you are used to traveling to Vegas for this content. I can't help you with that one, but for the atmosphere fix yourselves a cocktail before reading. Here goes:
like in years before, a lot going on here, I decided to focus on 4 new types of instances.
Instances for high performance graphic workloads, such as those in game streaming, animation, and video rendering, aiming to improve performance and reduce cost for graphics-intensive workloads.
C6gn instances deliver up to 100 Gbps network bandwidth, up to 38 Gbps Amazon Elastic Block Store (EBS is a storage service designed for use with EC2 for both throughput and transaction intensive workloads at any scale) bandwidth, up to 40% higher packet processing performance, and up to 40% better price/performance vs x86-based network optimized instances.
AWS started with generalized compute and had its team build a chip on top of Arm that it calls Graviton, which first manifested itself in A1 instances for scale-out workloads - things like web-tier workloads. it caught on.
Graviton2, AWS’ second version of the Graviton chip, which it claims gives better price performance than the most recent generations of the x86 processors from other providers and C6gn is an example for some this improvement's implications.
3. EC2 R5b Instances for 3x Higher EBS Performance - R5 instances are designed for memory-intensive applications such as high-performance databases, distributed web scale in-memory caches, in-memory databases, real time big data analytics, and other enterprise applications.
4. MAC EC2 Instances Long awaited solution for developers creating and customizing macOS, iOS, iPadOS, tvOS, and watchOS Apps. Not cheap instances, but this is just 1st generation. Knowing AWS, that will change in due time.
One of the top 5 uses of AWS users and where it delivered 2 dramatic announcements, one specifically answering a clear ask by its users and one answering its users need before they iterated it into a specific request.
5. S3 Replication - Adds the ability to replicate data from one source bucket to multiple destination buckets, completing the AZ replication solution, with replication across regions and even across accounts, with user defined rules. S3 replication may be the most overdue announcement, in any case it is one of the most anticipated ones. Up till now users needed to create their own work flow or Lambda function to replicate data across buckets, if they were not in the same Availability zone.
6. S3 Storage Lens – Big News here – A first cloud Object Storage Analytics Solution, that allows users to understand, analyze, and optimize storage usage with over 30 metrics through an interactive dashboard aggregating data from one or more organizational accounts, across all regions, with actionable recommendations, already appearing in users S3 management console.
This has the potential to change how users use S3, find anomalies, waste, duplicities, etc. Customer obsession wise, AWS nailed it on this one. If you're an AWS customer, you need to be checking this out regularly.
It's become a tradition over the past 5 years to expect a lot from re:invent here. This year met the bar.
7. Habana Gaudi-based EC2 Instances – Built specifically for ML training, Intel chips by Habana, formerly on of Intel Capital's portfolio companies. H1 ‘21. The Gaudi accelerators are designed for training deep learning models for workloads that include natural language processing, object detection and machine learning training, classification, recommendation and personalization. Up to eight Habana Gaudi accelerators will power each EC2 ML instance.
8. AWS Trainium – New ML training chip, support all major frameworks and the Neuron SDK, H2 ‘21.
The company promises that it can offer higher performance than any of its competitors in the cloud, with support for TensorFlow, PyTorch and MXNet. It will be available as EC2 instances and inside Amazon SageMaker, the company’s machine learning platform. New instances based on these custom chips will launch next year.
The main arguments for these custom chips are speed and cost. AWS promises 30% higher throughput and 45% lower cost-per-inference compared to the standard AWS GPU instances.
9. AWS Panorama - AWS enhances its ML tool kit with a machine learning Appliance and Software Development Kit (SDK) focused to aid companies to bring computer vision (CV) to on-premises cameras to make predictions locally with high accuracy and low latency.
10. Amazon Redshift ML - A solution which makes it possible for data warehouse users to create, train, and deploy machine learning models using SQL commands. This is a real time saver and makes a lot of sense. I would bet it will need to improve and become more robust with time, but sounds like a promising start.
I'm a fan of serverless ever since Lambda's Inception and think AWS hit the nail on the head with these two:
11. Lambda Billing - Thus far, AWS rounded your usage of Lambda up to the nearest 100 ms spot. It is now down to 1 ms granularity. This means REALLY pay for exactly what you use. This will make Serverless even more cost effective, if people didn’t get the hint by now.
12. Lambda function memory increase - Lamnbda functions can now support up to 10GB memory allocation – 3x over former allocation – so Lambda now supports a wider range of uses requiring up to 6 virtual CPUs. Existing uses can run faster and be billed the same, or roughly the same. Lambda users can allocate up to 10 GB of memory to a Lambda function. This is more than a 3x increase compared to previous limits. So you can get things done faster, and pay the same – More memory, less time.
A lot here as well, but I chose one.
13. Amazon Monitron – AWS sensors for Industrial IoT to be scattered around sites. Sample vibrations and temperatures to a central hub connected to AWS , detecting anomalies for predictive maintenance application.
Thus far I think Monitron is one the most interesting plays in IoT by AWS, as it targets a direct vertical of IoT, making it available for self deployment by end users.
- Hybrid/Multi Cloud Architecture
It is far from trivial that AWS, THE cloud company, would make maverick technological plays in this direction. But the matter of fact is, it has and it kept making them in 2020.
14. Amazon Outpost per server – AWS’ vSphere powered OnPrem solution is now available per single server, instead of a full server reck up until now. Billed per server, paid monthly.
If I had to guess, while most enterprises tell AWS they are indeed willing to start experimenting with Hybrid solutions (87% of them actually, according to Flexera), they are just not ready to do it at the pace AWS originally expected and they are over budget as a rule on cloud expenditure as is, almost by 25%.
15. BubbleFish – New feature allowing applications written for MS SQL work with Aurora without changing anything to the application, connects legacy databases to cloud applications.
Aurora is Amazon' cloud based MySQL compatible relational database - AWS Bubblefish is basically - IMHO - the AWS plan of attack on Microsoft SQL and Oracle DB.
16 and 17. ECS anywhere/ EQS Anywhere – AWS’ managed container service and Kubernetes service are now available on any cloud or on-prem environment (NK: WoW!)
Amazon ECS is AWS’ fully managed container orchestration service and EKS is its service allowing to start, run, scale, patch provision and update Kubernetes Applications. Thus far you could run both on AWS cloud or Outpost.
Running both anywhere, as far as implications to enterprises adoption, this supports Hybrid Cloud adoption and Multi Cloud. Amazon EKS and ECS will run on-premises customers' existing infrastructure, or with their other cloud architecture. This a significant shift in the cloud provider's hybrid container management strategy that may also reduce enterprise container adoption and deployment costs.
For context AWS has more than 100,000 active customers using ECS, and billions of compute hours on EKS run every week on AWS, according to Andy Jassy's keynote.
- AI Vericals
AI is very important to AWS and it never settles on just providing building blocks for AI based solutions, it also provides some solutions it has built based on its own building blocks and you can assume these have been piloted on Amazon.com or one or more of the many AWS key customers, like Netflix. This is both a way for AWS to broaden its offering and showcase what you can build with its technology.
So this isn't something we did not see before, HOWEVER, I do not recall so many AI Vertical announcements around a single solution suite, like what AWS did this year with their Call Center management solution, Amazon Connect.
And finally, a personal favorite in AI Verticals, really interested in seeing how well it works and how companies will put it to use:
24. QuickSight Q - NLP based service allowing users to ask questions in natural language and receive business answers.
So there it is, far from complete and I invite you to add more in the comments and share with other readers/contributors.
As an investor, and former Amazonian, I follow AWS re:invent each year. It's a great way to understand trends, competition, partnering opportunities and ways for my portfolio companies to get more done, more effectively.