Software Performance - A Shape Not a Number, by Kay Ousterhout
tl;dr: Percentiles are not enough, you need histograms, e.g.:
Really neat talk about finding performance issues in really complicated systems, like the many-service worlds that many companies have now.
Moving from 1 to N regions - an open retrospective by Andrew Bloomgarden
tl;dr: How to introduce new tooling carefully, and what can go wrong
This sounds like things I’ve experienced personally, except done with more resources.
Isolation without Containers by Tyler McMullen
builds on mozilla’s craneshift jit compiler, except it’s not a jit.
also, fastly compiles vcl (and maybe soon webassembly) from customers to native code and runs it on their cdn (and every request?!)
tl;dr: Weird, charming, and entertaining. And weird.
A nice example of making your talk a story.
- Algorithms for DNA Data Storage (Ruthie Nachmany)
- ‘Scalable Anomaly Detection (with Zero Machine Learning)’ by Arthur Gonigberg-6UwcqiNsZ8U.mp4