[Python-Dev] Re: a serious threat to 2.3's speed?

[Michael Hudson]

"Delaney, Timothy C (Timothy)" <tdelaney at avaya.com> writes:

>> From: Fredrik Lundh
>> 
>> key features contributing to this is true gc, call-site 
>> caching, and "traditional"
>> (C-style) argument passing for common cases.
>
> Did you collect stats on how much garbage collection actually
> occurred during the tests? How did it compare to stock CPython?
>
> One of the big flaws in benchmarking early versions of Java of
> course was that most tests didn't end up invoking the garbage
> collector - hence showing an artificially good result that could not
> be observed in real-world usage.

If you read the pytte1 pages you can find /F talking about how
accidentally disabling the GC wrecked performance.  My suspicion is
that Python apps allocate and deallocate so many objects that not
having a GC around leads to ferociously cache hostile behaviour.

Cheers,
mwh

-- 
  Famous remarks are very seldom quoted correctly.
                                                    -- Simeon Strunsky