Python Performance: Past, Present and Future
summary
Survey of Python implementations and why many JIT projects stop: several research efforts target big speedups but fail to deploy; Unladen Swallow aimed 5x, PyPy shows ~2–10x on some workloads and ~4x in benchmarks, and most teams solve hot spots by rewriting in other languages. Practical wins available now: Cython, Numba, multiprocessing with shared memory in multiprocessing and faster pickle, plus ongoing work on a new stable C API (handles), tracing GC and subinterpreters to attack the GIL and C‑API limitations.
AI-written summary · about
chapters (5)
- 0:00Introduction
- 1:51Past optimization attempts
- 9:27Present performance solutions
- 19:54Future of Python performance
- 43:45Q&A
related talks
58mPython performance Past, Present, Future
22mPyPy and the future of the Python ecosystem
57mPython and PyPy performance (not) for dummies
01h 03mA tour of CPython's runtime
26mPython Performance at Scale - Making Python Faster at Instagram
59mMyths and fairy tales around Python performance
22mPyPy in production
22mPyPy hands-on