by Dan Roncadin | Chief Consultant at Taos
AWS re:Invent, one of the most highly anticipated cloud events of the year, did not disappoint.
Every year AWS continues to put out a more comprehensive set of services, and sets the bar even higher. They’ve gone from being a provider of infrastructure and foundational components, to offering managed services that make it easier to run things like databases and cache clusters, now to offering artificial intelligence and machine learning capabilities that would be extremely difficult and costly for clients to replicate on their own infrastructure.
I am curious what my fellow cloud enthusiasts think about the key highlights and notable changes.
1) A different approach to key announcements: Did it focus on the right things?
Game-changing announcements that would have been front and center in the keynote presentation of previous years were relegated to being revealed on the AWS blog, which was incredibly active even above its normal daily torrent of updates. It’s a tough situation, as the amount of updates being continually released could make for keynotes that go on for hours, as it was Werner’s already went well past its schedule. In any case, as a user of the service, it’s clear the focus is on pitching the brand new capabilities of the platform, not enhancements to existing offerings.
2) The new “M” offering & bare-metal servers: Compelling?
The families of virtual machine instances on the EC2 service were expanded with a new “M” offering, the M5 instance. M family instances are the most widely used instance types as they strike a great balance between compute and memory with a 1:4 ratio of cores to GB. These are offered at a 5% discount to the M4 family but have nearly 15% faster CPU performance.
Price/performance increases such as this used to be headline news, but they’re just part of the normal course of business now for AWS.
For the first time, AWS is offering true bare-metal servers. This will be attractive not only to users who were previously using “dedicated tenancy” for their instances, but also to people wanting to run large servers packed with containers avoiding the hypervisor overhead, latency-sensitive workloads, and software with strict machine-licensing requirements.
3) AWS Guard Duty: Taking over more of the burden of Cloud Security?
Guard Duty is a new service that makes fantastic use of both built in platform capabilities (like CloudTrail and VPC Flow logs, DNS Query Logs) plus new machine learning abilities to provide intelligent threat detection by analyzing events to alert on potential threats (such as if an EC2 instance in your account starts contacting a known bitcoin mining pool IP address). Previously this capability was largely absent from the platform, as the AWS Shared Security model had Amazon providing security “of the cloud” while the customer was responsible for security “in the cloud”. This new service has AWS moving up the stack and taking more responsibility for the security that was previously left up to customers to solution. I’m certain a few of the security companies exhibiting at re:Invent weren’t thrilled by this competing offering.
4) Artificial Intelligence and Machine Learning: Is this the next killer app for cloud platforms?
Artificial Intelligence and Machine Learning were the highlights of the conference this year, with amazing services such as Amazon Transcribe – which will use machine learning and AWS’ vast computing capability to offer fast scalable transcription of audio and video recordings into text. Not to be outdone on the meta front, Amazon introduced SageMaker which helps you train machine learning models by itself using machine learning.
AWS upped their hardware giveaway to attendees this year, moving from the Echo Dot of 2016 up to a DeepLens device this year. DeepLens is a combination camera enabled linux machine with the ability to run machine learning models locally. Now everyone can write their on hot dog, not hot dog app.
Machine Learning takes perfect advantage of massively scalable compute resources that you can also turn off once the training is complete. AWS and the other cloud providers are really differentiating their offering from what is plausible to accomplish with traditional infrastructure and limited capital.
5) Serverless – Is AWS extending their lead?
Lambda is the leading serverless cloud platform, but new capabilities announced really show how serious AWS is about moving the area forward. There are the basics like increased memory and language support for Golang, but the more interesting parts come in the database arena. Serverless Aurora takes server less beyond functional programming into hosted services. Relational data layer persistency is the missing link in serverless architectures where current capabilities cause continued reliance on always-on dedicated instances and services like RDS.