The speed of PyPy's Python
==============================

Summary:

The first part of the talk will cover PyPy's speed achievements resulting 
from the last year's work on the Just-In-Time Compiler.  I'll present 
and discuss a number of benchmarks and compare against other Python-speed 
projects. I'll also cover the basics of how the JIT works and what 
sort of programs it can greatly speedup (and which ones it can't).

Description:

PyPy's JIT has been in development for some years now.  More recently
it became more practical and begins to be `faster than Psyco`_ on some
examples.  This is the status in September 2009 and currently it's fast-paced
development so we expect to have even more interesting results for
February 2010, worth discussing and comparing.  If possible
a 45 minute talk would be cool due to the "explaining the JIT"
bit of the talk. 

Proposed talk agenda:

* Show various benchmarking data for comparison of different
  Python interpreters (5 minutes)

* Describe what is JIT with some examples (10 minutes)

* A bit of introduction into how PyPy's JIT work (it's a generated tracing JIT,
  like the tracemonkey JS interpreter, unlike Psyco or JVM JITs). (5 minutes)

* Tracing JITs focus on often executed loops. Which kind of programs
  will the PyPy JIT speed-up, which constructs we know it is currently poor at.
  (15 minutes)

* Future plans for PyPy, especially JIT-wise.
  (5 minutes)

* Q&A (5 minutes)

.. _`faster than Psyco`: http://morepypy.blogspot.com/2009/09/first-results-of-jit.html

The main part of the talk would focus on a detailed explanation what
information JIT can use (at the time of talk) or at all and what
constructs it'll be poor at.

Also, the talk will be focused on intermediate audience, hence prolonged
timeline for explanation of what is JIT and how it works.