Several members of the ScaleVP team spent last week in Las Vegas at Amazon’s annual re:Invent conference. The environment was predictably chaotic with tens of thousands of developers, infrastructure company leaders, and investors hustling between meetings, sessions, and keynotes.
Throughout the week, the Company made dozens of new product announcements (an exhaustive list can be found here), with a particular emphasis on new machine learning capabilities, and what should come as no surprise, additional compute resources. At a high level, Amazon has long been known as the most formidable leader in the cloud computing arena and will continue to add more features to its arsenal in this respect. With that said, this year the Company’s leadership also demonstrated a concerted effort to extend Amazon’s machine learning platform and capabilities so as not to give up ground to Google and Microsoft. We should all expect more developments here in due course.
Below, we’ve attached some notes and commentary on some of the announcements that we found especially noteworthy.
From Keynote with Andy Jassy, CEO of AWS:
Notes on the Business:
- AWS is now an $18B revenue run rate business growing 42% YoY with “millions” of active customers
- AWS has achieved 44.1% market share in the infrastructure as a service market; Microsoft is the next largest vendor with 7.7%
- The AWS team will launch an estimated 1,300 new capabilities in 2017
- 63% of Kubernetes that runs in the cloud runs on AWS (despite having been born out of Google)
- AWS Lambda (function as a service) has grown 300% YoY (Amazon Aurora, specifically, is the fastest growing single service in the history of AWS with > 2.5x YoY growth)
Key Announcements:
Amazon Elastic Container Service for Kubernetes (EKS):
- This was highly anticipated
- Think of as a managed Kubernetes service that will sit on top of AWS
- Will allow developers to deploy and orchestrate containers across multiple availability zones and will be compatible with hybrid environments
- Will automatically run upgrades and patches and will be integrated with many AWS features such as Cloudtrail and Elastic Beanstalk
- In order to make all of this possible, Amazon will be leaning on Heptio and Tigera, which were highlighted by Company leaders on several occasions
- Developers will now be able to easily, build, train, and deploy machine learning models on AWS
- Can choose to use pre-built algorithms for common problems and train algorithms with one click (no longer requires that companies have a separate team to do so)
- Leverages a methodology called hyper-parameter optimization (“HPO”), which spins up multiple copies of ML algorithms and then layers on top additional ML to optimize the model
- Can be deployed in one-click across multiple availability zones
- Will handle node failures and perform routine maintenance and health checks
- Because of compute resources, pre-built algorithms will run anywhere between 3x – 10x faster than other providers
- Will allow developers to run containers on ECS (Amazon Elastic Container Services) or EKS (Elastic Container Service for Kubernetes) without having to manage or choose servers and clusters of EC2 instances
- Fargate is like EC2 but it will provider developers with containers instead of virtual machines
- Will permit developers to run containers at the task level and at the server level
- Will permit on demand auto-scaling of databases for applications with unpredictable or cyclical workloads
- In other words, developers will no longer be required to provision databases instances at all, and can automatically scale capacity up and down as needed
- Developers will be able to pay by the second and only for database capacity that is used
- This is now the only cloud database that will provide on continuous backup on demand
- Allows for point in time restore up to the second for the last 35 days
- Can now backup hundreds of TB’s of data with no effect on performance
- Developers can now pull out select data and objects they need using standard SQL expressions (as opposed to extracting all objects / data in bulk)
- Devs can leverage a new API to select and retrieve data within objects and accelerate any application that processes a subset of object data in S3
- Will improve data access performance by up to 400%
Amazon Transcribe, Translate, and Comprehend:
- Automatic conversion of speech into accurate, grammatically correct text
- Will run in multiple languages – planning to start with English, but more languages will be released in the coming weeks / months
- Has support for telephony /audio (can also recognize multiple speakers)
- Users can add custom vocabularies and idioms that will be understood over time
- Amazon Translate will enable users to translate text into languages of choice
- Amazon Comprehend will allow users discover valuable insights from text (entities, key phrases, language, as well as sentiment analysis)
- Will allow users to scan millions of documents to build topic models
- More consumer / developer focused, but very interesting use case of computer vision capabilities
- Deep Lense is an HD video camera with on-board compute optimized for deep learning
- Think of as a video camera that can help to build, train, and tune machine learning models
- Integrates with AWS SageMaker and AWS Lambda
- Can begin to run inferences within 10 minutes of unboxing
- Example use cases might include a) users are sent a text message when their dog jumps on couch at home or b) users could program their garage door to open only when it sees a certain license plate
- Will run computer vision model directly on the device and can identify objects and even emotions
- Will cost $249 on Amazon
Keynote with Werner Vogels, CTO of AWS:
Key Themes:
- Voice will be the next major disruption in computing – Amazon intends to become a leader in this delivery
- Transition towards human-centric interfaces (e.g. voice) both in the home and office
- The increasing need for automation in security
- The rise of micro services – independent scaling, deployments, security / permissions, and improved fault isolation
Key Announcements:
- A fully managed service for controlling multiple Alexa devices at work
- Companies are able to provision and manage devices, users, and skills
- First use case is reshaping conference rooms; “Alexa, start the meeting”
- Will integrate with O365 and G-suite, on prem exchange (calendars, schedules, meetings), Ring Central, Salesforce, Concur, and Splunk (among others)
- Worth noting that the The Wynn hotel has committed to adding an Alexa to each room in its luxury collection to enable voice-based process automation
- A fully managed cloud IDE (integrated development environment) for writing, running, and debugging code
- Rendered in the browser with a full terminal and availability in many languages (Ruby, Python, and Java)
- Provides collaborative editing capabilities (similar to Google Docs) for devs
- Key tenants will be faster deployments, integrations, real-time pair programming, and broad debugging support for Lambda
