AWS Summit London 2018 Review

AWS Summit is a global series of conferences run by Amazon to showcase Amazon Web Services. Here, code-along labs are hosted, speakers proselytise on juicy new bits and general best practice in the Cloud, Amazon Partners tout their wares, Amazon offer certification — all this wrapped up in the usual developer conference atmosphere. This year's UK event was hosted at the cavernous ExCel Centre in Docklands, East London, and I joined 12,000 other people and took the first morning train out of the regions to join the party.

Getting to the venue was trivial on the Tube and DLR, and after registration — a half-hour penned in like cattle in theme park-style crowd management furniture — we were ready to rock. To be fair, I've suffered worse queueing at other festivals. Like Glastonbury. The queueing is definitely worse at Glastonbury.

Coffee in hand, I ambled through the vendor exhibition swag-pit for the keynote.

Introduction and Keynote

Gavin Jackson, the MD for AWS UK and Ireland, warmed up the packed, 4,000-seat auditorium by thanking us all for coming and giving an overview of the recent trends in AWS. Most impressive was the fact that AWS's year-on-year growth is now nudging fifty percent, and that this rate of adoption is actually increasing. It's pretty clear that we're a long way off peak Cloud.

After highlighting some newly launched features, not least Enterprise Contract for AWS Marketplace, allowing vastly simplified licensing for proprietary software, he briefly interviewed Peter Virk, Director of Connected Car & Future Technology at Jaguar Land Rover on-stage. Peter described how AWS services had been instrumental in the development of JLR's Connected Car project, by which a driver can start their car and various of its functions remotely, which is apparently desirable. Finally Gavin introduced Dr. Werner Vogels, Amazon CTO, to outline some of the new services AWS has started offering over the last few months. For anyone that follows the break-neck pace at which AWS evolves, it won't surprise you that this was a long keynote.

First up was the Cloud9 web-based IDE. This impresses by allowing the running of Lambda functions (Amazon's Function-as-a-Service offering), and integrating closely with AWS Code* ('Code-star' — CodeCommit, CodeDeploy etc.) family of services. It also has some novel collaborative features, allowing a near-pair-programming development model to be used by distributed teams. I'd love to have the opportunity to use this in anger at some point.

Machine learning features on AWS have expanded significantly recently, with Amazon SageMaker introduced as a tool for the building of a usable model from pre-canned algorithms from your organisation's data. Lots of companies are using this to powerful effect. But as it is a goal of Amazon to make machine learning technology available to all developers, not just ML specialists, there are a range of services that sit atop SageMaker to make leveraging this power simpler; the Rekognition image processing service outlined as an example of this. There was a brief presentation by Babylon Health, who have been deploying chatbot-style apps for diagnosing health conditions using various ML techniques, with an impressive 30% of Rwanda's population signed up.

Databases and storage is another big area of progress, with a recent expansion of Aurora to now have a PostgreSQL-compatible option, in addition to the MySQL support it offered previously. The Amazon take is that standard PostgreSQL falls short of enterprise-grade performance and reliability, and the rewriting of this database's backend ameliorates this, not least by automatically replicating all data via six replicas across three Availability Zones. S3 Select and Glacier Select were also mentioned, supporting the retrieval of partial S3 (Amazon's object storage solution) objects via an SQL-like interface.

New services were described for security, such as Amazon Inspector for the automatic detection of vulnerabilities; and for microservices, such as the EKS preview allowing deployment of containers across a Kubernetes cluster, and Fargate for the deployment of containers without the need to manage their underlying hosts. More on Fargate later.

There were brief presentations from and HSBC and... crikey, the time! This has overrun — I better hotfoot it out of here and get to the next session...

"Creating and Publishing AR and VR Apps with Amazon Sumerian"

So this one was unlikely to have much immediate effect on either my career or productivity at work, but it seemed pretty cool, so...

Sumerian was announced last year as a tool for building and publishing Virtual Reality and Augmented Reality interfaces, and is currently in preview. Named, I was pleased to learn, in homage to the use of the Sumerian language in Neal Stephenson's novel Snow Crash, it offers a web-based graphical 3D IDE with which to compose VR/AR scenes, including a 'Publish' wizard for making them immediately publicly available. Ben Moore, lead on the AWS Sumerian program, ran us through it.

The IDE works in the latest browsers and provides a simple interface for the composition of these scenes. There's a rich asset library, as Amazon have actually been importing wireframes of items that their not-insignificant retail arm has access to, such as furniture and the like. The IDE is designed to be usable with no coding experience necessary, although Javascript is available to script complex behaviours. Drag-drop objects onto the scene, and behaviours can generally be assigned to objects via drag-dropping them around a graphical state machine and configuring them there.

Then when you're done, click go, and you get a CloudFront (Amazon's CDN service) URL for the viewing of your scene in a normal browser (latest versions of Chrome and Firefox.) Your Android phone will probably handle it nicely — if you've got an Oculus Rift plugged in, or one of multiple other devices, so much the better. Pricing is by scene storage and scene data transfer.

Awesome stuff. I wish there were more demand for Augmented Reality from the financial services sector; then I'd be using the heck out of it...

"Design with Ops in Mind"

Outside the next session I noticed the "Grab & Go" lunch stalls. They had a range of sandwiches, snacks and drinks that you could just help yourself to and stuff in a paper bag on your way past. As I usually dine al desko, and had steeled myself for a big queue at a retail establishment for lunch, this suited me just fine, so a win here.

The previous session had been in one of the conference rooms upstairs, suitably shielded from the rest of the event; but this one was in one of the "Theatres" — really just giant curtain boxes in the main exhibition space alongside the food and the vendor stalls. These were fine, but if I had to nit-pick I'd mention that the sound insulation between the theatres was poor and you could always hear three presentations at once.

It wasn't unduly difficult however making out Alex Livingstone and Liga Viskinte, Senior Technical Account Managers at AWS, as they tag-teamed their spiel. This really was a high-level overview of various AWS services right the way through the delivery pipeline, the majority of which I've encountered before, but some it was good to see for the first time. There were demonstrations of CodeCommit and CodeDeploy for Continuous Delivery, and X-Ray for tracing across a whole microservices architecture, and... crikey, the time! This has overrun — I better hotfoot it out of here and get to the next session...

"Serverless Authentication and Authorisation for your APIs on AWS"

I had been really keen to go to this session but, by the time I arrived, the doors were locked and the population density inside the room was described by one front-of-house staff member as "dangerous". Frankly with everything that's going on in the world I at least expected some safety here, so I just went across the hall to the nearest available talk.

"Let Me Graph That For You"

So here I found myself at a lecture I'd almost randomly selected about whose subject I knew absolutely nothing, but it turned out to be really interesting. Ian Robinson (not our Ian Robinson!), Specialist Solutions Architect, gave a demonstration of Neptune, a new AWS service currently in preview.

Neptune is a graph database supporting two graph models: Property Graph via its query language Apache TinkerPop Gremlin (I am not making this up), and RDF (Resource Description Framework - specified by W3C) via its query language SPARQL.

A graph database is a natural choice for a large, highly interconnected dataset supporting the fast retrieval of relationships necessary for various machine learning workloads, and Amazon's offer here supports ACID transactions with immediate consistency, and all the friendly database features you would expect from AWS such as read replicas, multi-AZ deployment and encryption at rest.

A series of case studies were then presented by Semantic Integration Ltd., for example demonstrating how a machine learning algorithm powered by a graph database proved invaluable in automatically associating a hotel's facilities, location etc. with a journalist's review of that hotel, with no input necessary from the journalist other than to confirm the property was matched successfully.

"Deep Dive into AWS Fargate"

I'm sure Fargate is the future, so this was the single talk I was most looking forward to. The organisers clearly thought this would be a common opinion as it was scheduled in the massive Auditorium, but in the event the venue was mostly empty.

Fargate is a new service from AWS that allows the provisioning of containers without having to worry about setting up the underlying hosts' infrastructure. This is a huge step change in application deployment from Amazon: although up until now ECS (Elastic Container Service) has allowed for container workloads to be deployed and scaled across a cluster, the set-up and maintenance of the EC2 host nodes comprising that cluster has still been the user's responsibility.

Abby Fuller, Senior Technical Evangelist, described in detail how Fargate has changed all that. You've now got two options when you want to deploy ECS tasks: first, do you want to use the new EKS Kubernetes preview as a cluster type, or stick with 'classic' ECS (the overloaded use of "ECS" here was noted and apologised for)? And second: do you want to use the 'classic' EC2 ECS backend, or Fargate?

Using Fargate is simpler to get up and running. It's currently only supported in four regions, though luckily for us in Western Europe, Ireland is one of them. But using EC2 provides greater flexibility; for example, it isn't possible to "exec in" to a Fargate container.

Fargate permits only a single network type for your containers: the custom type "awsvpc", under which each container gets its own Elastic Network Interface with its own private IP. Public IPs can also be assigned. As an added bonus for Fargate, there's now a bewildering plethora of CLI tools available for maintaining it. There's not only aws-cli, and ecs-cli (useful if you want to use Docker Compose) — there's now also the third-party Fargate CLI and Coldbrew CLI tools too.

"Come Out from Behind your Firewall"

And finally, an introduction to the maddeningly capitalised world of "DevSecOps". Chris Schrom and Bill Baldwin one-twoed a discussion of how security has evolved as architectures have moved from traditional on-prem to Cloud-native. A common pattern discussed involved using SNS (Simple Notification Service), invoked by a Lambda, to send notification emails as a result of CloudTrail, CloudWatch or VPC Flow Log events following aggregation in S3.

This was then followed up with a breakdown of the Amazon GuardDuty service. GuardDuty automates the monitoring / response / remediation cycle to "intelligently" detect threats. An example demonstrated was an alert triggered by an EC2 instance querying a domain known to host malware.

"Networking Drinks"

It had been a long day. I was now feeling pretty exhausted, or at least extremely antisocial, and so necking a glass of red and a couple of reasonable vol-au-vents was about all the networking I could manage before heading back. To conclude:


  • Huge and varied programme
  • Free; free lunch; free booze
  • Some rockstar speakers (e.g. Abby Fuller)
  • Great venue
  • Great phone app


All minor, but in the interest of balance:

  • Registration could have been smoother
  • Overrunning sessions — I have sympathy with the speakers because this must be so hard to gauge, but it is frustrating when an overrun causes you to miss your next session
  • Over-/under-capacity sessions — Surely it should be possible to provisionally state an interest in advance to better gauge likely attendance? You can already rate sessions via the app! The technology is here!

So all in all, a great day! I wouldn't hestitate to recommend this to any developer with an interest in DevOps and Cloud services. Unsurprisingly given the breadth of Amazon's offering, the scope of the subject material was enormous, and there should be something here for everyone. It was generally very well run and I certainly enjoyed it immensely.

Thanks Amazon!