Extra 8: Scalability! But at What COST?

• GraphX paper
  • lots of cores 128

• repeated look at each label in the graph
  • flows through

• ran a single core in Rust and it was fast!

• with better algorithms! - Even faster

• scalability fetishised by community

• at 16 cores it is the same performance at a single core

• 1, 2 cores
  • very very fast! micro-seconds

• still high performance on a single laptop!

• When you are focued - high performance compiled languages, caches, not JVM

• step 1. get 100x on your computer
• step 2. then 100x from 100x computers
  • but that's expensive!