Keynote: Performance tuning
summary
Optimizing compilers and machine learning cut real costs not just buzzwords: Google applies ML to data-center controls and saves 40% cooling 15% overall energy, Twitter’s VM work plus autotuning reduces allocations by 7% memory per tweet and drops p99 latency about 28% yielding roughly 80% fewer machines for the same load. Datacenters already use ~2% of global electricity, Bitcoin mining costs ~78 TWh/year, Microsoft charges $15 per metric ton carbon internally, and simple client features like Twitter Light and data saver cut network traffic.
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:28Who am I
- 3:27Compiler optimization
- 5:55Data centers and electricity
- 11:39Twitter example
- 13:08Bitcoin example
- 14:09Electricity consumption
- 17:32Bitcoin
- 22:24Data reduction
related talks
01h 20mGreen Software Architecture - Dos Don'ts and Some Surprises
30mFighting Climate Change by Building Sustainable Software
01h 23mCreating Better Computer Systems by Focusing on Sustainability
14mIs your Code Net zero?
34mCarbon Footprint–Aware Software Development
38mArchitecting for a Greener Future
42mEnergy-Efficient Software Architecture for Developers
46mMaking Software & Data Architectures Sustainable