Good Morning!

Twitch’s president put out a rather stunning blog post about why they’re doing a 50/50 revenue share with streamers in most cases. The reason I’m mentioning it here is the quote "Lastly, we have to talk about the cost of our service. Delivering high definition, low latency, always available live video to nearly every corner of the world is expensive. Using the published rates from Amazon Web Services’ Interactive Video Service (IVS) — which is essentially Twitch video — live video costs for a 100 CCU streamer who streams 200 hours a month are more than $1000 per month. "

This is not even slightly credible. As it turns out, I ran an IVS economic analysis in June of 2020 using (funnily enough) Twitch’s traffic numbers. The conclusion was that there was no possible way that Twitch was paying anything even in the same universe as the published IVS prices; if they had been they’d have been almost 25% of AWS’s annual revenue, and been spending ten times their acquisition cost on AWS every year.

Though if someone at Twitch wants to argue otherwise, I would be THRILLED to help you shave many billion dollars a year off of your AWS bill; please reach out.

From the Community

Don’t confuse infrastructure resilience with data resilience. AWS helps with one of those, Clumio helps with the other. Protect your data against cyberthreats, hackers, and bugs. Discover Clumio!

I found a blog post contrasting AWS VPC peering vs Transit Gateway; I’m going to be doing my own exploration soon just because I don’t fully understand the nuances here. Plus I want to include Cloud WAN in the comparison.

This incident report about some ingestion delays demonstrates why Honeycomb is the archetype of incident transparency that we should all aspire to.

This visualization about how S3 deletes 1 billion objects uses Athena and Rust to explore something very interesting.

A comparison of AWS Compute Options in 2022 is handy; the story does keep evolving for most workloads.

Aidan Steele is extremely back on his bullshit, which to be explicitly clear are his words and not mine. He’s found a horrifying way to avoid part of the Managed NAT Gateway tax and let two instances in different private subnets communicate with one another without the existence of a third party relay. It’s both amazing and horrifying.

Forsooth, AWS services explained for you in Victorian AWS.

I’m interviewing to be The Next AWS CMO.


Last Week In AWS: Connecting All William-Nilliam

Last Week In AWS: The Next AWS CMO: Corey Quinn

Last Week In AWS: The Swole Architected Framework

Screaming in the Cloud: Azul and the Current State of the Java Ecosystem with Scott Sellers

Screaming in the Cloud: How Data Discovery is Changing the Game with Shinji Kim

Choice Cuts

Push your code when it’s ready, launch it when you are ready, and most importantly: log off you’re ready and not worry about your coworkers calling you at 3AM and screaming you the songs of their people, specifically "Rage in D Minor." Take control of your launches with AWS AppConfig Feature Flags, and stop causing problems for other people on purpose.

Amazon CloudWatch Evidently adds support for Client Side Evaluations for AWS Lambda (powered by AWS AppConfig) – CloudWatch Evidently and AppConfig (two services that effectively do what feels like the same thing) integrate to do more of that same thing together.

Amazon Corretto 19 is now generally available – Joke’s on you if you thought that the first 18 Amazon Corretti had said all they had to say on the subject.

Amazon Route 53 announces support for DNS resource record set permissions – Your favorite database just got row-level access controls.

AWS Health Dashboard displays localized dates and times for Service Health events – This is a nice quality of life enhancement. I will also reiterate that whenever you’re doing backend systems or database timezone work, the one true timezone is UTC. Localize at the presentation level, never the data source level. Or your future self will hate you.

Using Prometheus to Avoid Disasters with Kubernetes CPU Limits – Let’s be very clear here: the scenario described is "you’re running Kubernetes, and then a second disaster strikes."

Build NFT metadata access control with Ethereum signatures and AWS Lambda authorizers – There was never a good time for nonsense like this, but "before the bottom fell out of the NFT craze" would have been a better one.

How Informatica® Cloud Data Governance and Catalog uses Amazon Neptune for knowledge graphs – Yeah I don’t care about Neptune; how exactly did Informatica bully AWS into including a registered trademark symbol next to their name? A whole bunch of companies would have preferred AWS do that, but Informatica actually pulled it off.

See how is using AWS to drive innovation at massive scale – Amazon uses Amazon to become more Amazon.

Detect population variance of endangered species using Amazon Rekognition – I’m borderline worried that some of the Machine Learning® services are so focused on being provably correct that if this analyzer gets a wrong result it’ll dispatch poachers to make the real world fit the model.

O’Reilly Book on Observability Engineering — Get Yours Free from Honeycomb!
Manage complex cloud-native systems, improve customer experiences, and build & run better software using Honeycomb. Get your FREE copy of our new O’Reilly book and register for our Authors’ Cut Series to discuss key concepts.

How Amazon Search reduced ML inference costs by 85% with AWS Inferentia – Then it reduced them even further by returning just a list of house brands and sponsored ad placements instead of organic search results, by all appearences.

Copy existing AWS CloudTrail trails events to a AWS CloudTrail Lake event data store – This is, to be blunt, freaking awesome. I don’t do particularly well with "data manipulation and management," so CloudTrail Lake has been a godsend for me–but the big challenge I’ve had with it is that it only contains events that occurred since you enabled it. This enhancement lets you import historical data into it, meaning I don’t have to stumble my way through more annoying ways of querying historical data. Not every feature release is for every customer, but this one sure is for me.

Announcing AWS Config Compliance Scores for conformance pack – Gamify compliance, where the high score is you get to keep your job, and second place includes a prison term.

AWS announces low-to-no cost security services for federal political campaigns and committees – This is a good thing. "Cybersecurity" is one of those things I believe people deserve that transcends partisanship.

How AWS supports the 10th anniversary of National Voter Registration Day with digital tools for civic engagement – No snark here. This is awesome; it’s AWS at its absolute best.

How one nonprofit commemorates enduring sibling relationships with AWS – This puts the "bro" into "techbro" for sure. I would appreciate a moment of silence for the approximately 40 ribald comments I could have made and somehow decided not to…

Nara Space uses AWS to improve satellite image quality up to three times with deep learning – But it costs as much as transferring the images from the satellite twenty times, so really, who’s winning? AWS. <strike>The House</strike> AWS always wins.

New report: The economic impact of the new AWS Region in the Middle East – The UAE region apparently costs significantly less than their New Zealand region, as per their previous economic impact report. This is presumably due to the famous New Zealand hobbit tax.


Getting to the cloud is challenging enough for many organizations, especially maintaining security, resiliency, cost control, and agility. Things break, configurations drift, technology advances, and organizations need to evolve. How can you get to the cloud faster and ensure you have the right team in place to maintain success over time? Work with a partner who gets it – Logicworks combines the cloud expertise and platform automation to customize solutions to meet your unique requirements. Get started by chatting with a cloud specialist today.

This Lambda Live Tuner is handy for overcoming a shortcoming in the Lambda Power Tuner: specifically, some Lambdas aren’t able to be run repeatedly and idempotently, so this modifies real-world Lambda request configurations and then returns a result.

… and that’s what happened Last Week in AWS.

Newsletter Footer

Sign up for Last Week in AWS

Stay up to date on the latest AWS news, opinions, and tools, all lovingly sprinkled with a bit of snark.

"*" indicates required fields

This field is for validation purposes and should be left unchanged.
Sponsor Icon Footer

Sponsor a Newsletter Issue

Reach over 30,000 discerning engineers, managers, and enthusiasts who actually care about the state of Amazon’s cloud ecosystems.