If you need to speed up your Python program there are several possible approaches, with varying degrees of effort required. Here is a (likely incomplete) overview from late 2008:

1. Rewrite or Parallelize Your Python Code

Optimize, tune, or replace algorithms — or parallelize execution using:

  • Hadoop or Disco — Open-source implementations of MapReduce
  • Parallel Python
  • Message Passing Interface (MPI) — Frequently used for numerical programming
  • Bulk Synchronous Parallel (BSP)
  • RPyC — RPC for distributed and parallel programming
  • Twisted — Networking libraries for distributed/parallel programming
  • Profiling Tools — Python standard profile / cProfile
  • Threading or Microthreads (Stackless)

2. Use a Tool That Speeds Up Code Mostly Unchanged

  • Psyco — Just-in-time compiler (probably the easiest first approach to try)
  • Pyrex — Write and compile Python with C-style data structures
  • Cython — Compiling Python-like code with optional static types to C extensions
  • PyJS — Compile a large subset of Python to JavaScript (more relevant for client-side Ajax than server-side compute)
  • RPython — Compile a restricted subset of Python to native code (part of the PyPy project)
  • PyStream / GPULib — GPU acceleration libraries for Python
  • Shedskin — Compile a large subset of Python to C++ (benchmark showing 196× math speedup)

3. Replace Parts of Python with Another Language

  • SWIG (Simplified Wrapper and Interface Generator) — Call C/C++ from Python
  • F2PY (Fortran to Python Interface Generator) — Call Fortran from Python
  • llvm-py — Generate and compile directly to native machine code via LLVM
  • CorePy — Write assembly code directly in Python
  • SciPy Weave / PyInline — Inline C/C++ snippets inside Python
  • Boost.Python — Interoperability between C++ and Python
  • Cinpy — Blend C code with Python using ctypes and runtime compilation via tcc