How Go is Making us Faster
summary
MediaMath moves a monolithic API to microservices to handle 2.5 million req/s with ~40 ms decision budgets, 120 TB/day across nine datacenters. Teams prototype production Go services in a weekend, ship static binaries, and hit real numbers: a dynamic creative service at 95% < 10ms on two boxes per pop, an auth service at 1 ms 95th, and a Plex multiplexer >100k reqs/node. Uses goroutines/channels, gofmt, go test (+race), pprof and vet; warns about GC pauses, no generics and ad hoc dependency vendoring.
AI-written summary · about
chapters (31)
- 0:00Intro
- 1:56What is this talk about
- 2:10Context
- 3:19Microservices
- 4:00New problems
- 6:06Easy to solve
- 8:12Awesome standard library
- 9:20A vibrant community
- 10:04The tool chain
- 11:06Go test
- 12:31Race detector
- 13:02Go Vet
- 13:30Productivity
- 14:24Speed
- 14:50Go vs C
- 15:27Go vs Pearl
- 15:44Go vs Java
- 16:39Go Services
- 18:10Performance
- 19:00Go Tests
- 19:39Sampling Profiler
- 20:22Garbage Collection
- 21:27Go Max Procs
- 22:24Channel is Magic
- 23:14Mutex Lock
- 23:32Empty interfaces
- 24:07Runtime type checking
- 24:22De defer
- 24:54De unwind
- 25:17GCC
- 26:14Things we dont like
related talks
45mGo, Microservices and all the joy!
46mSimplifying Scalable Cloud Application Development with Go
31mSuccessful Go for microservices architecture
31mBuilding a microservices architecture with Go
01h 03mGo & Microservices
53mGo at Mainframe Scale
01h 02mBuilding Microservice Architectures with Go
30mHigh Performance Systems in Go