home ‣ Psyco - effortless, significant speedup of Python programs   login

Summary: using psyco is an easy way to get a 9x speedup for real-life Python program.

I just wrote a Python program that took quite a while to complete. It was reading a 600 MB *.sql file (uncompressed Wikipedia database dump, parsing it to extract some fields and writing the result to a text file for easier processing in the future). While it's not a production code it is important for me. Currently Python is my favorite language for writing non-GUI code because of ease of development and rich libraries.

The only sore point is its speed, or lack of it but I will always trade CPU cycles for programming cycles. There are efforts to speed up Python code and one of them is psyco. I've heard about it before but never tried in practice so I decided to give the latest version (1.2) a try. Setup (on Windows XP, Python 2.3.2) was a breeze. All it takes to use psyco is to add 2 lines to your program. What about the results? I achieved 9.67 speed up which is great. When, for benchmarking purposes, I limited the processing to only 10.000 (out of around 360.000) records, the processing time dropped from 301.19 seconds to 31.12 seconds. That's a big win and makes a real difference when processing the whole file. Go psyco!

Standard benchmark warning: "there are lies, damn lies and benchmarks". My app is different than yours so it might get better or worse speedup from psyco. Having said that, he app in question is non-trivial, real-life app (i.e. it's not a micro-benchmark designed to test a specific feature) so I think it's fair to say that psyco can give a serious speed up to a large class of python apps. It seems to be stable and is extremely easy to use.