Saltar al contenido ForoHackers

Performance engineering: start here

Ingeniería de rendimiento
#885 · Ophuscado · · Denunciar

Performance engineering is the practice of making software use less time, memory, energy or bandwidth, guided by measurement rather than guesses. Most programs spend nearly all their time in a small part of the code, so the skill is finding that part, understanding why it is slow and changing the right thing.

Key concepts

  • Measure first: profile before changing anything; intuition about bottlenecks is often wrong.
  • Algorithms and data structures: a better complexity class beats any amount of tuning.
  • Memory hierarchy: caches mean access patterns matter; contiguous, predictable data is fast.
  • Concurrency and parallelism: using several cores helps only when work divides and contention is low.
  • Latency and throughput: how long one request takes, versus how many are done per second.

Vocabulary

  • Profiler: a tool that shows where a program spends time or memory.
  • Hot path: the code run most often or longest.
  • Cache miss: an access that has to go to slower memory.
  • Amdahl's law: the limit that the serial part of a task puts on speed-up from parallelism.

What to study, in order

  1. Complexity analysis and how to read a profile and a flame graph.
  2. Profile a slow program of your own, find its top cost, fix it and re-measure.
  3. Rewrite a loop for cache-friendly access and measure the difference.
  4. Learn the system's counters and tracing tools to see CPU, memory, disk and network use.
  5. Parallelise a task, then find where contention or waiting limits the gain.

Related tags

#performance-evaluation #algorithms #computer-architecture #web-performance #observability

Inicia sesión o únete para responder.