AWS re:Invent 2019 Andy Jessy KeyNote

New Instance Type: Ifn1, M6g, R6g, C6g

Inf1 Intstance equipped with AWS Inferentia is optimized for machine learning operations and costs. Compared with G4 Instance with the same level of computing performance, the cost is reduced by
40%.

Equipped with the latest Arm-based AWS Graviton2 Processor, the performance is around 40% higher than the current x86-based M5 / R5 / C5, which will be available globally in 2020 Q1.

• SPECjvm® 2008: +43% (estimated)
• SPEC CPU® 2017 integer: +44% (estimated)
• SPEC CPU 2017 floating point: +24% (estimated)
• HTTPS load balancing with Nginx: +24%
• Memcached: +43% performance, at lower latency
• X.264 video encoding: +26%
• EDA simulation with Cadence Xcellium: +54%

Reference source to：Graviton2-Powered General Purpose, Compute-Optimized, & Memory-Optimized EC2 Instances

Fargate for EKS !!

When Fargate was released from re:Invent 2017, it mentioned that it would support both ECS & EKS container services. After waiting for two years, it finally officially supported EKS in GA!

Currently,eksctl, a widely used tool for deploying EKS Cluster, also releases version updates in time, supporting Fargate as a deployment unit for pods:

eksctl create cluster --name demo-newsblog --region us-west-2 --fargate

In the EKS Cluster Console, you can also use the Fargate Profile to declare the Fargate resources needed when pods are deployed.

Reference source to：Amazon EKS on AWS Fargate Now Generally Available

Accelerated Site-to-Site VPN Connections

Now you can freely choose whether to enable Accelerate on Site-to-Site VPN connection to speed up the access speed of VPN connection.

The accelerated VPN connection uses the AWS Global Accelerator released last year, allowing On-Premises device points to be routed to AWS edge location (PoPs), thereby replacing relatively remote VPC VGW endpoints, obtaining lower connection latency, and accelerating the overall connection line speed. Shortening the connection distance can also reduce the possibility of VPN Connection being disconnected due to the network disconnection of the Public Internet.

Reference source to：Accelerated Site-to-Site VPN Connections

Multicast on Transit Gateways

The communication methods of Broadcast and Multicast were not supported on AWS. If users have related requirements in the past, they need to use EC2 as an overlay network solution on VPC to virtualize Multicast.

The multicast connection between VPC Subnets is now supported through Transit Gateways.

Multicast domain：

• Create a Multicast network environment
• Select the Transit Gateway to be the Multicast Router
• Associate VPC Subnets to Multicast domain

Multicast group

• Create Source as Multicast Sender or Multicast Receiver
• Multicast group IP as IPv4 class D：224.0.0.0/4
• Group members are EC2 Instance ENIs

Multicast source

• EC2 Instance ENI sending Multicast

Multicast group member

• EC2 Instance ENIs cluster receiving Multicast

Reference source to：Amazon Transit Gateway

Amazon Managed Apache Cassandra Service (MCS)

In addition to Amazon DynamoDB, NoSQL Database Solutions of Amazon DocumentDB (with MongoDB compatibility) released last year by re:Invent, this year it also released Amazon Managed Apache Cassandra Service (MCS) with Apache Cassandra as the managed object.

It is a serverless service like DynamoDB. Users only need to pay for actual usage, and automatically extend AutoScaling according to load conditions to ensure the performance of Tables. In use, they also interact with Tables through Cassandra Query Language (CQL).

This is good news for users who are not easy to migrate to DynamoDB / MongoDB.

Now it has Open Preview Regions：

• US East (N. Virginia)
• US East (Ohio)
• Europe (Stockholm)
• Asia Pacific (Singapore)
• Asia Pacific (Tokyo)

Reference source to：Amazon Managed Apache Cassandra Service (MCS)

S3 Access Points

When the user’s same S3 Bucket has many permission-controlled bucket policies, it is very annoying for the maintenance staff. When we only want to modify one of the access permissions, we are afraid that it will move Other projects, so S3 Access Points is a good solution. We can add access points for different objects or applications, and let these objects be used individually for individual control.

Reference source to：Easily Manage Shared Data Sets with Amazon S3 Access Points

Sagemaker Studio

The most significant update is a major change in Sagemaker in the field of machine learning.

Announced an integrated machine learning IDE, which contains: Web Integrated Development Environment (IDE) for machine learning (ML). You must have IAM permissions before using this service. This service has AWS’s own notebooks called Sagemaker Notebooks (Preview Phase), which makes it easy to create and share Jupyter Notebooks. The second one is SageMaker Experiments, which can make a one-time comparison for each training job to assess which training is most effective and quickly deploy this model. When we train models, we will inevitably encounter problems. SageMaker Debugger will automatically check the trained models and collect data for analysis to provide real-time notifications and recommendations. SageMaker Model Monitor can view the quality deviations (deployment results) of deployed models and receiving alerts. Finally, in machine learning, selecting algorithms is a very difficult problem, especially choosing and mastering the best model that can solve the problem. Machine learning algorithms usually require a large number of training parameters. These parameters need to be set usually after multiple corrections and training to get the most suitable value to reduce the accuracy of the model. Users need to call the API once or with a few clicks in Amazon SageMaker Studio. SageMaker Autopilot first checks the dataset and creates many data preprocessing steps, machine learning algorithms and hyperparameters to run tests. This combination can then be used to train the pipeline and deploy it to a real-time endpoint or batch process.

This figure is a simplified diagram of the functional correspondence in the Sagemaker service after this update.

Reference source to：ついにSageMekerの統合環境が登場！「SageMaker Studio」が発表されました #reinvent

Amazon Fraud Detector

This AI updated service Fraud Detector is a fully managed service that identifies potential fraudulent online activities such as online payment scams and the creation of fake accounts. Leverage machine learning and 20 years of fraud detection expertise from AWS and Amazon.com to automatically identify potential fraud activities in milliseconds.

Reference source to：Amazon Fraud Detector FAQs

Amazon CodeGuru

This new AI service is a fully managed code inspection service that can identify serious flaws in code and deviations from Java-based AWS best practices. It currently supports GitHub and CodeCommit.

Reference source to：Amazon CodeGuru FAQs

Amazon Kendra

Amazon Kendra is a machine-learning and easy-to-use enterprise search service that allows developers to add search capabilities to their applications so end users can discover information stored in a large amount of content across the company. This includes data from manuals, research reports, FAQs, HR documents, customer service guides, and can be found in various systems such as file systems, websites, Box, DropBox, Salesforce, SharePoint, Amazon S3, and more. When entering a question, the service will use machine learning algorithms to understand the context and return the most relevant results, whether it is an exact answer or the entire document. For example, The user can ask: “What is the cash reward on the company credit card? “, Amazon Kendra will search for the relevant document and return something like” a specific answer like 2% “.

Reference source to：Amazon Kendra

AWS Wavelength

AWS Wavelength can connect AWS services to 5G networks, allowing developers to deliver applications to various mobile devices and end-users with ultra-low latency performance of fewer than 10 milliseconds. For future use in games, live streaming or AR / VR will achieve better results.

E-commerce and regions currently supported:

• Verizon in North America
• Vodafone in Europe
• SK Telecom in South Korea
• NTT Docomo, and KDDI in Japan

Reference source to：AWS Wavelength

Contact Lens for Amazon Connect

This is a service that supports Amazon Connect through machine learning, enabling call center executives and analysts to understand the content, emotions, and trends of their customer conversations, so as to identify key customer feedback and improve the customer experience. The customer service center receives a large number of daily customer information, resulting in millions of hours of call history.The company wants to be able to search across all phones to identify issues, common topics and opportunities for agent guidance. They can use existing contact center analysis products, but these tools are expensive, provide call records slowly, and lack the required recording accuracy. It has been difficult to quickly detect customer issues and provide actionable performance feedback to their agents. Existing tools can not provide real-time analysis，that is, customer service personnel can not judge the customer’s current mood in real-time. Now, you can use Contact Lens for Amazon Connect to determine the current mood of your customers in real-time, and quickly search the customer’s frequently-created topics and a number of functions that are beneficial to customer service personnel.

Reference source to：Contact Lens for Amazon Connect