AWS re:Invent, the annual Amazon extravaganza in Las Vegas, is on the horizon. As a technology and cloud geek I am excited to be attending for the first time in person. I found a great introductory event guide here by Mark Nunnikhoven, with useful hints and tips for new attendees like me. My focus will be to check for updates to AWS IoT services, especially integrations with juicy AI and ML service updates. There will certainly be new and exciting announcements in the keynotes. I will make some educated guesses on what these might be. The re:Invent twitter account also publishes regular updates and tips about the event. Photos used here are by our co-founder Neil Miles at the 2018 event; the photo above was taken from the front row. I am told by our co-founder and CTO, Robin Meehan (who will be presenting on 4th Dec), that it’s really odd sitting right in the very front, as the stage is massive and weirdly still quite a long way away, like watching Ed Sheeran at Wembley. This year, ambassadors like our own Phil Basford (also presenting on 2nd December) will be in the front row. We have certainly come a long way very quickly since we first attended in 2017.
Recent Updates
But first, AWS are “clearing out the cupboard” with a spate of recent announcements, mainly about improving and extending their core IaaS, PaaS and SaaS offerings. Here are some things that have caught our attention among the many updates in the last few weeks, preceding re:Invent.
*New* – Several IoT announcements on AWS Iot Day the day after publishing this blog! These include the ability to do secure tunnelling and configuring your own domain and SSL Server Certificate for the IoT Endpoint, which is great for our highly security-conscious customers.
In the AI/ML space, Amazon Comprehend has added support for many more languages, including Chinese and Arabic, which is relevant to our customers in the Middle East region. Amazon Transcribe is also extending the supported languages at a fast rate, now supporting Dutch, Farsi and Portuguese for example. We are building bespoke interactive voice-interfaces using combinations of these services especially where keyboard use is difficult.
The recent announcement of a Data Science SDK, to orchestrate pipelines around Amazon SageMaker takes away some of the effort required to integrate our models into production flows. It uses Step Functions, which provide simple ways to encode workflows across Cloud Services.
We like stats, and recently I heard on the AWS TechChat podcast that WordPress powers 34% of all websites world-wide. Anyone who’s built a simple website has probably used WordPress and it is hugely configurable via plugins. AWS has released an Amazon Polly plug-in, together with Amazon Translate to translate the content. Hosting on AWS also enables CloudFront distribution of your website. As with all hosted applications, secure access to your AWS back-end services using IAM roles.
Savings Plans are the next phase of AWS’s cost efficiency services like Reserved Instances (RIs) and using budget calculators, scheduling, etc. Typically, a one-year Reserved Instance (RI) saves 30%-50% over on-demand cost when you run the instance 100% of the time. Reserving for longer periods can save a little more and paying up-front a little more again. RI purchasing has evolved a lot but with usage varying so much depending on how you distribute your application across regions, availability zones let alone how you auto-scale it gets complex and many customers don’t take advantage as much as they should. Savings Plans give discounts similar to RIs if you commit to use so much compute per hour over the plan period (1 to 3 years). This gives simplicity and flexibility, even giving recommendations based on usage in the early days of a production launch service.
Compute Savings plans can save up to 72% costs and covers instances regardless of region and includes compute used in containers, including AWS Fargate (AWS’s managed container service). This could be a key factor in decisions to migrate your workloads to Fargate and as websites scale moving away perhaps from Serverless Lambda apart from event-driven functions. When you’re regularly training ML models your compute can be very ‘peaky’ so it is important to consider a way to optimise your costs using tools like cost manager.
CloudWatch Region View allows sharing of CloudWatch data across accounts. This is really useful to those of us having to regularly switch between accounts. CloudWatch is how we monitor services, gathering logs, calculating metrics, alerting and many other functions. Best architecture practice is to centralise CloudWatch logs and provide dashboard views as well as the Insights feature to help find problems across services. It is highly flexible and configurable so cross-account, cross-region dashboards (secured using the IAM policies) simplifies administration. We have also started using the new Service Lens feature to build service maps. There is also an anomaly detection feature in CloudWatch for detecting unusual and unexpected events in log files. As Machine Learning matures we expect to see AWS including these kinds of ML-driven features in more and more of their traditional services.
Another fairly recent example that we’re using in solutions is Glue FindMatches which integrates ML models for finding matching terms across data sources, which can be used for example to match customers across multiple different back-end systems. I anticipate that such usages of ML will be one of the features mentioned most often in re:Invent sessions, even if they are not part of the core “ML Family” of services.
Cloud providers are always looking to take away the “undifferentiated heavy lifting” of doing common tasks and using AI/ML to analyse data is almost becoming utility so it takes away the barrier to starting out on the ML journey. As always, if you want to really find an edge and make the best use of these services and make the best decisions for your business then we can help.
CloudFormation resource import allows you to bring existing resources into CloudFormation. Although we always build our environments using automation (either CloudFormation or, increasingly, Terraform) there is certainly experimentation via console in the early days of projects so the ability to reverse-engineer to CF as a starting point is valuable.
A vital tool for DevOps is the AWS Command Line Interface (CLI) which allows access to services from your client, using a terminal. CLI v2 is coming soon (still in developer preview) has SSO out the box, meaning automatic short-term credential management, which enables developers to switch between accounts and roles seamlessly without refreshing credentials. Check it out if you’re using DevOps CLI functions on AWS.
IoT Device Management has been enhanced with Fleet Level Metrics and improved ways to query across devices. This will be able to easily identify things like overheating batteries which are in the top percentile of temperature in vehicles and gather more information to trigger preventative actions. There are also some very handy operational improvements in managing IAM roles and the cross-instance restore enhancement to Redshift.
Crystal ball time
What announcements do I expect at this year’s re:Invent?
re:Invent 2018 was big and 2019 will be even bigger
I expect further, closer integrations between services, many of which will involve links into SageMaker Machine Learning tasks, like the CloudWatch anomaly detection and FindMatches functions within Glue. There may be some announcements regarding Big Data, including more updates to RedShift and integrations with QuickSight becoming even easier.
I expect further developments of newer technology-based services like Neptune (in the growing graph database space) and QLDB/Managed Blockchain as the synergies between these technologies and AI/ML are being realised, for example in Graph Analytics.
Enhancements to the core machine learning services will be of most interest to us of course. I expect support for more and more powerful algorithms in SageMaker and higher instance memory sizes to run training jobs. There are also lots of development and code pipeline announcements, perhaps we can see some inroads into Agile development, managing stories, etc within the service.
There are many ways to extract, transform and load data in AWS. I think that data orchestration is going to be enhanced to bring together the various ways of managing data flows into and out of Data Lakes and other data stores with easier definitions and flow management. One interesting idea I heard recently was to take data out of structured databases and “mix it up a bit” to see if the imposed structure of systems had hidden connections.
Conversational interfaces are at the top of the Gartner Hype cycle for 2019 and set to reach maturity within 2 years. Given the aforementioned increases in language support I expect a convergence in end to end conversational services enhancing Lex Bots (Alexa skills type services in other words) that can be integrated into solutions AND incorporate translation services.
No recent new regions or availability zones (nor even edge locations), which maybe means there could be some coming. We have already made good use of the Bahrain region which was one of the most recent launches and there are still lots of growth areas where global customers may wish to deploy services for regulatory or other business reasons. South America and Africa look ripe for further connectivity and the APAC regions are growing rapidly. Further global rollout of services seems inevitable. Price reductions, offering savings to loyal customers may be further promoted in competition with other Cloud providers.
And finally, I am looking forward to having some fun and meeting our lovely customers and AWS colleagues who are attending the event. Playing with Deep Racer anyone? They are even putting on the new Star Wars movie Rise of Skywalker so maybe I will join our own Ambassador Phil to catch that. If you cannot attend in person, remember that you can Live Stream the event.