Skip navigation to main content

AWS re:Invent event - Andy Jassy Keynote

KCOM’s own Wayne Stallwood is at the AWS re:Invent event in Las Vegas right now. Here are his thoughts on some of yesterday’s key announcements.

A great series of launches were delivered by Andy Jassy today, I think as with other years it’s possible to see where Amazon are focused. Below are the ones which really grabbed my attention.

The new Machine Learning category in the AWS Marketplace simplifies the commercialisation of ML algorithms to the point that I think it’s just opened up a whole new market segment.

This fundamentally changes the accessibility to ML models that can solve real world problems. At launch the marketplace is already populated with products that can help with breast cancer prediction, loan risk prediction, vehicle model recognition or cyber security attack detection.

Each of these would have required significant effort to develop, the same problems being repeatedly solved by different customers. Creating an open market for these solutions can only intensify their development and quality.

Machine Learning for the masses sounds a bit like science fiction until it happens. I think we’ve only just landed.


Back in the reality of enterprise IT, Amazon has launched new services to help us with some of those problems those of us specialised in cloud migration often wish we didn’t have.

EFS is great for Linux environments, and now we have FSx for Windows File Server, offering us a fully managed SMB file service with Active Directory integration. Certainly, in the case of a migration project it’s very likely that there will be a requirement for network file storage so this is most welcome.

Managing Lustre isn’t easy, Amazon FSx for Lustre offers a managed service with the sort of performance spec that would be very hard for even us to build ourselves. KCOM customers who are operating HPC clusters should be very excited by this.


One of the challenges with DynamoDB is sizing, in what sometimes feels like an anti-pattern for an AWS managed service, capacity had to be provisioned ahead of demand, which ultimately meant you had to provision for peak.

DynamoDB on-demand gives us a very welcome flexible scaling and billing model, this will be great for peaky and unpredictable workloads and I think generally makes DynamoDB significantly more aligned to the rest of Amazon’s Serverless components.

The provisioned scale model is still available of course, it’s even possible to switch Dynamo Tables between models. Perhaps operating in the cheaper provisioned capacity model for operating periods within your service where the database workload is well understood and then switching to on-demand mode during periods of uncertainty.


AWS DeepRacer looks like a fun way to lean and develop skills around Sagemaker RI (reinforcement learning). Sadly following a pattern started by the AWS IoT button, and cloned by the AWS Storage Gateway hardware appliance, it looks like the UK is in for a long wait as won’t ship it to us.

All of this and we still have the keynote tomorrow to look forward to.  Stay tuned!