Performance with and without the garbage collector

[dieter]

Steven D'Aprano <steve at pearwood.info> writes:
> ...
> Is anyone able to demonstrate a replicable performance impact due to garbage
> collection?

I have had some experience with the performance impacts of garbage collection
-- not completely replicable but rather frequently visible.

Huge Zope instance (with around 2 GB heap usage; millions of objects);
a badly designed (external) component
that created thousands of temporary objects
during searches. What we could observe: occasionally, our Zope got
unresponsive for about 1 minute; a monitoring software typically
told us that at least on of Zope's worker threads was in the external
component mentioned above.
The analysis finally revealed: creating and releasing thousands
of short lived temporary objects frequently called for a
garbage collection. The garbage collection prevented
all other Python work while running -- and this caused outages in
the order of a minute.