Home > AWS, re:Invent > AWS re:Invent 2018: Werner Vogels Keynote Thoughts

AWS re:Invent 2018: Werner Vogels Keynote Thoughts

November 29th, 2018


Thanks to Eric Hammond for the great pic!

Some people say Werner is still delivering the long keynote from last year and we’re just coming in to see another instalment! Andy Jassy yesterday may be talking to the executives, today’s the turn of Werner to talk more technical and give us some of his considerable wisdom (you do know Werner isn’t just some “standard” tech CTO but one of the world’s recognised academic authorities on distributed systems now putting his brain into practical action at AWS).


Werner talks tech but also threads a about developing a new architecture for cloud use. Sure, you can just migrate your existing workloads to AWS but that’s going to cost you and generally keep your current complexity but now running somewhere else. Werner always advocates to truly take advantage of cloud you need a new architecture. His job is to sell you this vision so you take the time and effort to refactor your applications for the cloud to achieve a better business outcome and of course pay AWS rather than go somewhere else.

This keynote was billed as the Serverless Keynote


Werner started with the Oracle bashing immediate my mentioning his worst day at Amazon which was December 12 2004 when an Oracle database bug affected customers getting orders ready for Christmas. Werner went onto describe limiting your blast radius with cell based architectures, cell based units so only a unit would fail, not your whole system.

Relational databases are not designed for the cloud which was a link back to my Tim Bray talk yesterday (AWS re:Invent 2018: Inside AWS: Technology Choices for Modern Applications – SRV305) where he said they do not use relational databases if at all possible at Amazon, preferring DynamoDB or Amazon QLDB.


Werner talked about Amazon Aurora which was built as a scale-out distributed architecture database for the cloud. 6 replicas for much better availability. We went through quorum architecture and how they can be made efficient. Onto a history lesson and explanation of DynamoDB and how it is performance at scale.

Then more information on S3 delivered by the VP, Mai-Lan Tomsen Bukovec. Lots of detail on how availability and recovery after failure are built into the system at scale.


Werner then went onto his best day, 1 November 2018 this year when they switch off their largest Oracle database, why so much focus on rubbishing Oracle. I really wonder why this is so personal from a number of people speaking at the conference, is it a message for CxOs to trust AWS to be as good/better than Oracle at running DBs. Surely it can’t be a dig at Oracle Cloud which AWS must see as irrelevant. I wonder if Larry nicked Andy’s sun lounger on holiday one year? Also mentioned Redshift 3.5 faster than 6 months ago.

Fender then came on stage talking about the history of the guitar and how they are using AWS services to engage new customers better. They even gave Werner a guitar.




Holly Mesrobian, Lambda Director of Engineering came on stage to talk about Firecracker and how small, efficient and optimal it is.

Werner back on stage talking about coding environments, IDEs, Languages, Programming Models and Workflow, everyone has their own opinion on what to use.

Some announcement, a whole string of serverless ones:

  • AWS ToolKit for PyCharm, Intellij and VSCode. Local debugging for Lambda
  • Ruby support for Lambda. Custom Runtimes so you can bring your own languages
  • Lambda Layers: avoid duplicated code, link libraries centrally so you don’t have to load the libraries on each invocation. This looks very interesting for 3rd parties to be able ot link a security scanner for example once and get all functions to use it which will make it more efficient.
  • Nested Applications using Serverless Application Repository: Compose applications architected from reusable building blocks
  • Step Functions with service integrations, connect 8 AWS functions together without writing code.
  • WebSocket support for APi Gateway: for real-time two way applications. No reason not to have API Gateway fronting your entire business, very important, API Gateway is great although could do with a better pricing model.
  • ALB Support for Lambda: I had thought this was already done.
  • Amazon Managed Streaming for Kafka: fully managed highly available Kafka Service

National Australia Bank on stage to talk about a number of projects they’re running on AWS from eFX to their Data Hub. They’re training more than 3000. I contract for a competitor to them so was good to see a company with similar challenges being bold with AWS.

Well Architected


The Well Architected Review is a very important part of AWS, making sure you’re using the best AWS services for the job and covering your availability, security and scale.

Werner then announced what I think is one of the most important announcements, the Well Architected Review Tool, a service which looks across all of your environments to “Measure and improve your architecture using Well Architected Best Practices”. How great is that, are you doing key rotation, are you logging, are you securing this or that. AWS can see all your configuration across all your accounts and can deliver amazing insights and recommendations. This should not be underestimated. Cloud is hard, AWS is hard, So many choices, so much to configure despite a lot of it being taken care for you. You need something to check you’re on the right track. Architecture continually evolving and you generally only do an architectural review at the initial of a project. Image being able to continually do an architectural review of your environment as its running.

re: Play Party

Werner as is traditional announced the main act for the evening part which is 2014 returning



The keynote had a lot of technical detail. I do think it was perhaps a little too much. I know the point is to show the deep technical decisions Amazon is continually making to make their service easier, cheaper, more scalable etc.  However, part of the cloud deal is not having to worry about details. It took more than an hour for an announcements to roll in and would have been more useful IMO to look at new announcement use cases and see some in action.

Categories: AWS, re:Invent Tags: , , ,
Comments are closed.