re:Invent 2020 re:Cap (week 1)

By | 2020-12-05

We’re at the end of week one of the THREE WEEKS of AWS re:Invent. I have had a bit of time to digest the 1/3rd of the conference that has transpired and thought I’d put down my thoughts to paper (or keyboard as it were). The big event this week was the Andy Jassy keynote on Tuesday morning. Subsequent sessions have been going deeper into the details of said announcements. Here is a list of the announcements (along with some commentary from my perspective)

  • Habana Gaudi processor based EC2 instances built specially for ML
    • 40% better price/performance than current GPU based instances for ML workloads
    • Gaudi accelerators are designed for training deep learning models like NLP, object detection, and ML training/classification/recommendation/personalization
  • AWS Trainium
    • ML training chip custom designed by AWS to deliver the most cost effective training

Machine Learning is getting cheaper/better/faster all the time. If you are not looking into use cases for AI/ML to apply to your business, you are missing the boat.

A tacit acknowledgement that other computers exist on the planet that DO NOT belong to AWS. This will make multicloud efforts easier even though Jassy still did not say the word “multicloud”.

Lambda 1 ms billing – making the cheapest way to run workloads in the cloud even cheaper

  • AWS Proton
    • Managed application deployment service for container & serverless apps

DevOps tool for monitoring deployments & providing design templates to the dev teams

  • gp3 Volumes for EBS
    • 20% lower price point per GB
    • Scale IOPS and throughput without adding capacity

No more wasted disk space to get the IO you need from your EBS volumes

  • io2 Block Express
    • “SAN for the cloud”
    • 256K IOPS & 4000 MBps throughput
    • Max volume size 64 TB
    • Throughput scales 0.256 MB/s per provisioned IOPS, up to a max of 4000 MBps per vol

No more striping volumes for high IO workloads!

  • Amazon Aurora Serverless V2
    • Scale in fraction of seconds
    • 90% cost savings compared to provisioning for peak capacity (Yeah… but isn’t that what we are NOT supposed to do? 🙄)

  • Babelfish for Aurora PostgreSQL
    • Translation layer to understand commands written for MS SQL
    • Apps originally written for SQL can now work on Aurora “with fewer code changes”
  • AWS Glue Elastic Views
    • Combine, collate data from different data stores
    • Copies data from data store(s) to a target store
    • Use standard SQL to create your virtual table

Will play a key role in Data Analytics

No, not this Data

  • SageMaker Data Wrangler
    • Simplifies the process of data prep from weeks to minutes
    • Including data selection, cleansing, exploration, & visualization
    • Single visual interface
  • SageMaker Feature Store
    • Managed repo to store ML features
    • Tracks the metadata of the stored features for query purposes
  • SageMaker Pipeline
    • Purpose built CI/CD service for ML

Making that SkyNet future a certainty, one day at a time!

  • Amazon DevOps Guru
    • ML powered service to help improve an apps operational performance & availability
    • Automatically detect operational issues and recommend actions

This looks very cool; I am excited to learn more about this in the coming weeks

  • Amazon Connect Wisdom
    • Leverages ML against your internal KBs/wikis/FAQs to reduce time agents spend finding answers
  • Amazon Connect Customer Profiles
    • Automatically scans/matches customer records present a unified customer profile
    • Information such as name, phone number, and email address to help the agent identify the caller in real-time
    • Agents have all the customer information they need in a single place, so they can deliver more personalized customer service
  • Contact Lens for Amazon Connect
    • Enhances existing Contact Lens service
    • Now with real time sentiment analysis
    • Uses ML to detect customer experience issues during calls
  • Amazon Connect Task
    • Prioritize, assign, track, and automate contact center agent tasks
  • Amazon Connect Voice ID
    • real-time caller authentication

A bunch of call center tools to help you get that REALLY angry customer the help they need faster 🤣

  • Amazon Monitron uses ML to:
    • End-to-end service including sensors, and ML service
    • Detect abnormal machine behavior early
    • Implement predictive maintenance
    • Reduce unplanned downtime
  • Amazon Lookout for Equipment
    • For customers with existing equipment sensors
    • Uses data from your sensors to detect abnormal equipment behavior
    • Sends sensor data to AWS to build ML Model

Enhances your existing onsite cameras, which is pretty cool. Appliance means the site doesn’t need a constant/reliable internet connection, which is even cooler!

I am currently asking if AWS will put a 1U Outpost in my house 😊

  • 3 new AWS Local Zones
    • Boston
    • Houston
    • Miami
    • 12 more cities coming (all in the US)
  • AWS Wavelength:
    • 5G device infrastructure offering optimized for mobile edge computing applications
    • Now available in Las Vegas, San Francisco Bay Area, NYC, Atlanta, Dallas, DC, & Miami
    • Available soon in Korea, Tokyo, and London

That is it for the week one round-up of announcements. Two more weeks to go. I’m excited to see what the keynotes from Vogels, DeSantis, and Sivasubramanian will bring!

One thought on “re:Invent 2020 re:Cap (week 1)

  1. Dennis Faucher

    Wow, great write up. Thank you for saving me the days that it would have taken for me to find, and understand all of this. Bad news for NVIDIA with the hyperscalars building their own ML chips.

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.