The document summarizes announcements from AWS re:Invent 2016 related to compute, storage, artificial intelligence, serverless computing, databases, migration tools, and developer tools. Key announcements included new EC2 instance types, cost reductions, Elastic GPUs, AWS Batch for batch processing, Aurora PostgreSQL, Athena for analytics on S3 data, VMware on AWS, AWS X-Ray for tracing distributed applications, and expanded machine learning capabilities through services like Polly, Lex, and Rekognition as well as support for MXNet as an AI framework.
2. Agenda
My Goal:
Have an interactive
discussion about AWS
product strategy by
diving into re:Invent
announcements.
Agenda
• Attendance Trends
• Core themes
• Features
• Questions
8. Compute Enhancements
New Instances:
T2
R4
I3
C5
P2
F1
Cost reductions:
T2 / M4 / C4
The latest generation in a family is usually cheaper and
has better specs
Elastic GPUs (announced)
• Attach GPU’s like EBS
volumes
• Will be available to all 9
instance families
• Changes the economics of
GPU compute in the cloud
9. Making EC2 easier
VPS on AWS
Predictable pricing model
Instances run in separate VPC
VPC peering to integrate with existing
VPC based resources
10. Batch computing made easy
AWS Batch (preview)
• Fully-managed batch
processing at any scale
• Dynamic resource
provisioning, scaling, spot
fleet integration
• Priority queuing and
scheduling of jobs with
dependencies
Financial Services
Life Sciences
Digital Media
Use Cases
11. Security and Account Management
AWS Shield
• Managed DDoS protection
• Standard and ‘Advanced’
protection
• Advanced protection
includes WAF integration
(CloudFront/ELB)
AWS Organizations (Preview)
• Replacing consolidating
billing
• Provides ability to
• Create accounts
• Invite existing accounts
• Pay a single bill
12. AWS
Big Data
Artificial Intelligence Serverless
DevOps
Data
Compute / Networking / Storage
Security / Management
Migration and
Operations
Enablement
14. Clustered DB Engines
Making big data more attractive on AWS
Amazon Athena
• Serverless query service
that runs on top of S3
• Presto engine (ANSI-
SQL)
• JSON, CSV, Parquet, ORC
• Pay per query
• Fast performance
Data Lake
S3
RedShiftEMR
Athena
BI Tools
Data Scientists
ETL
15. Disrupting ‘enterprise grade’ databases
Aurora for PostgreSQL
(Preview)
• Full Postgres 9.6.1
compatibility
• Aurora benefits:
• Failover time < 30 sec
• 6 copies across 3 AZs
• Single-digit millisecond
replica lag
• At-rest and in-transit
encryption
SQL
Txns
Caching
StoreStore StoreStore
AZ 1 AZ 2
StoreStore
AZ 3
What makes Aurora Special?
20. VMware on AWS (Preview)
Fully managed VMware environment
on the AWS Cloud
Includes the VMware technologies
customers are currently running on-
premises
Combines the enterprise features of
VMware with the elasticity and security
of AWS
Components of VMWare Cloud on AWS
21. Configuration Management
Amazon EC2 Systems
Manager
• Tools for configuration
management of EC2 and on-
premises systems
• OS configuration
• Ad-hoc configuration
changes
• Run scheduled
administrative tasks
AWS OpsWorks for Chef
Automate
• Fully managed Chef server and a
suite of automation tools
• Compatible Chef cookbooks from
the Chef community
• Auto scaling groups for instance
registration and bootstrapping
22. Continuous Delivery
AWS CodeBuild
Fully managed build service that compiles
source code, runs tests, and produces
software packages that are ready to deploy
23. State Machines and ETL
AWS Step Functions
• Successor to Amazon
Simple Workflow Service
• Amazon States Language
Specification
• Lambda and ‘Activity’ Tasks
• States:
AWS Glue(pre-announced)
• Successor to AWS Data
Pipeline
• Fully managed ETL
• Source detection and
transform suggestions
• Built-in integration with
• Any JDBC-compliant
driver
24. Custom Scheduling Framework
• Enhanced CloudWatch events for
ECS state changes
• Container instances
• Tasks
• Allows for custom scheduling
• Cluster-state-service and
daemon-scheduler
ECS CLI
• For development
Containers
https://github.com/blox
• Tools and utilities to allow
extensibility of the EC2 Container
Service to better support emerging
requirements
26. Serverless Improvements
X-Ray (preview)
• Debug and trace end-to-end
requests in distributed apps
• Graphical topology of services
• Agent and API based integration
(Java/Node/.Net)
• Pricing will most likely be for trace
storage and retrieval
Serverless Application
Models
• Declarative CloudFormation
syntax for Serverless
• AWS::Serverless Primitives
• Lambda Functions
• DynamoDB Tables
• API Gateway
• (Swagger support)
• Infrastructure as code
32. Widening the AI
spectrum on
AWS
Artificial Intelligence Serverless
DevOps
Data
Compute / Networking / Storage
Security / Management
Migration and
Operations
Enablement
33. On one end of the spectrum:
AI Enabled Services
34. AI Enabled Services
Polly
• Natural sounding
speech
• Accurate text
processing
• SSML support
• 24 Languages, 47
voices
Lex
• Build Intelligent
chatbots
• Powered by same
technology as
Alexa
• Deploy to multiple
platforms
Rekognition
• Search, Verify and
Organize images
at scale
• Face Analysis,
Face Compare,
Face Recognition,
Object & Scene
detection
35. At the other end of the
spectrum:
‘DIY’
Deep Learning
37. AWS has picked its AI racehorse
MXNet (Mixed Network Models)
Programmability
Python, JS, R, Matlab
Portability
1000-layer network can fit inside 4GB
Performance
88% efficiency scaling up to 256 GPUs
New R4 Instances – The R4 instances are designed for today’s memory-intensive Business Intelligence, in-memory caching, and database applications and offer up to 488 GiB of memory. The R4 instances improve on the popular R3 instances with a larger L3 cache and higher memory speeds. On the network side, the R4 instances support up to 20 Gbps of ENA-powered network bandwidth when used within a Placement Group, along with 12 Gbps of dedicated throughput to EBS. Instances are available in six sizes, with up to 64 vCPUs and 488 GiB of memory.
Expanded T2 Instances – The T2 instances offer great performance for workloads that do not need to use the full CPU on a consistent basis. Our customers use them for general purpose workloads such as application servers, web servers, development environments, continuous integration servers, and small databases. We’ll be adding the t2.xlarge (16 GiB of memory) and the t2.2xlarge (32 GiB of memory). Like the existing T2 instances, the new sizes will be offer a generous amount of baseline performance (up to 4x that of the existing instances), along with the ability to burst to entire core when you need more compute power.
Coming in early 2017, I3 instances are ideal for the most demanding input/output (I/O) intensive relational databases, NoSQL databases, transactional systems, and analytics workloads. I3 instances have 15.2 TB of fast, low latency locally attached storage backed by Non Volatile Memory Express (NVMe) based SSDs to deliver up to nine times the IOPs as the previous generation with 3.3 million random IOPS at 4 KB block size with a total I/O throughput of 16 GB per sec.
Coming in early 2017, C5 instances include the next generation of Intel’s Xeon processors (code named Skylake) with AVX 512 and up to 72 vCPUs (twice that of previous generation compute-optimized instances) and 144 GiB of memory, making them the highest price to compute performance of any Amazon EC2 instance. C5 instances also feature new AWS hardware acceleration that delivers three times the Amazon EBS bandwidth of C4 instances. C5 instances are ideal for compute-intensive scientific modeling, financial operations, machine learning, and distributed analytics that require high performance for floating point calculations.
Amazon EC2 F1 is a compute instance with field programmable gate arrays (FPGAs) that you can program to create custom hardware accelerations for your application. F1 instances are easy to program and come with everything you need to develop, simulate, debug, and compile your hardware acceleration code, including an FPGA Developer AMI and Hardware Developer Kit (HDK).
Once your FPGA design is complete, you can register it as an Amazon FPGA Image (AFI), and deploy it to your F1 instance in just a few clicks. You can reuse your AFIs as many times, and across as many F1 instances as you like.
Amazon EC2 F1 instances are currently in preview in two different instance sizes that include up to eight FPGAs per instance. F1 instances include the latest 16 nm Xilinx UltraScale Plus FPGA. Each FPGA includes local 64 GiB DDR4 ECC protected memory, with a dedicated PCIe x16 connection. Each FPGA contains approximately 2.5 million logic elements and approximately 6,800 Digital Signal Processing (DSP) engines.
Quickly launch Virtual Private Servers on AWS
Include compute, SSD-based storage, data transfer, DNS management, and a static IP – for a low, predictable price.
Start simple, but use more AWS services as projects grows larger and more sophisticated over time.
For applications in fields like life sciences, activities like computational chemistry, clinical modelling, and genomic sequencing are enabled by batch computing.
There are many challenges involved in batch computing today even in the cloud: provisioning and maintaining a cluster, installing your batch processing software, managing dependencies between jobs, scaling, and job management.
AWS Batch allows developers access to cloud resources for batch processing without having to provision, manage, monitor, or maintain clusters.
Batch can automatically scale a cluster of managed servers to process a queue of jobs which can have their own priorities and complex dependencies.
AWS Batch takes care of the undifferentiated heavy lifting and allows you to run your container images and applications on a dynamically scaled set of EC2 instances.
This is in preview right now in us-east-1 and will roll out to other regions soon. More features are also on the way including the ability to leverage AWS Lambda to run batch jobs.
Cost:
Monthly Fee 3000 + .025 / .05 GB data transfer
Comprehensive protections from large and sophisticated attacks
Available on Elastic Load Balancing, Amazon CloudFront, Amazon Route 53
Advanced attack mitigation on Layer 3/ Layer 4 attacks
Includes AWS WAF for proactive rules
Near real-time notification via Amazon CloudWatch, with access to post-event analysis and investigation
Specialized support - 24X7 access to DDoS Response Team
AWS Bill protection against usage spikes due to a DDoS attack for ELB, CF and R53
At-rest and in-transit encryption – KMS integration for at rest and AES-256 for SSL.
Fun fact, if you had 100PB of data on site and fully utilized 4 10Gpbs DX ports, it would still take 215 days to complete the transfer
Sort of the final pieces to our code* services.
Chargted by the minutes
Different from simple workflow in that you don’t have to implement providers, the states json definition does that. Team that manages SWF and Step are the same.
Amazon States Language Specification – similar to a Business process markup language (BPML or BPAL). Similar to IBM, tibco, apian, etc
Step - you pay 2.5 cents per 1000 state transitions
cluster-state-service
The cluster-state-service consumes events from a stream of all changes to containers and instances across your Amazon ECS clusters, persists the events in a local data store, and provides APIs (e.g., search, filter, list, etc.) that enable you to query the state of your cluster so you can respond to changes in real-time. The cluster-state-service tracks your Amazon ECS cluster state locally, and manages any drift in state by periodically reconciling state with Amazon ECS.
daemon-scheduler
The daemon-scheduler is a scheduler that allows you to run exactly one task per host across all nodes in a cluste. The scheduler can be used as a reference for how to use the cluster-state-service to build custom scheduling logic, and we plan to add additional scheduling capabilities for different use cases.
ECS CLI does not support defining application load balancers or autoscaling groups
- Every one at the re:Invent hackathon used AWS Lambda in some form or the other
Amazon Polly supports Speech Synthesis Markup Language (SSML), a W3C standard, XML-based markup language for speech synthesis applications, and supports common SSML tags for phrasing, emphasis, and intonation. This flexibility helps you create lifelike speech that will attract and hold the attention of your audience.
Lex:
Automatic speech recognition (ASR) for converting speech to text, and natural language understanding (NLU) to recognize the intent of the text, to enable you to build applications with highly engaging user experiences and lifelike conversational interactions. With Amazon Lex, the same deep learning technologies that power Amazon Alexa are now available to any developer, enabling you to quickly and easily build sophisticated, natural language, conversational bots (“chatbots”).
Rekognition:
Amazon Rekognition is based on the same proven, highly scalable, deep learning technology developed by Amazon’s computer vision scientists to analyze billions of images daily for Prime Photos. Amazon Rekognition uses deep neural network models to detect and label thousands of objects and scenes in your images, and we are continually adding new labels and facial recognition features to the service.
Mention the deep learning ami and cloudformation template that autscales