I was going through the Redhat paper “What every programmer should know about memory”. Just collecting some topics that interested me when it comes to low-latency techniques. Not sure yet if I’ll do a deep dive on some or all of what I collect here. Paper review is definitely going to be done.

• Intro
• RAM (tbh not sure how useful for low-latency programming in my day2day)
  • SRAM v DRAM useful, remaining not too useful
  • Read-access protocol, DDR1->DDR2 might not be useful
• CPU (important)
  • program on disk, how does it get to ram and moves between caches
  • cache lines
  • Walk through memory hierarchy of computer
  • various levels of cache
  • l1 cache seperate between data and instructions
  • associativity - important
• virtualisation and virtual memory (important)
  • how does OS provide illusion that we have more resources to play with than available on hardware
  • address translation, mulitilevel page tables, tlb
• NUMA hardware (important)
  • topoplogy of CPU cores relative to memory
  • uma v numa
  • different data access patterns - how does it affect performance
• 6.4 concurrency optimisations
  • context swithces
  • sw v hw threads
  • oversubscribing
  • atomics e.g. cas
  • cost of threads
  • perf degradation