On Thu, 2008-06-12 at 20:57 +0200, Hrvoje Niksic wrote: > Eric Jonas <[EMAIL PROTECTED]> writes: > > > I've done some benchmarking while attempting to serialize my (large) > > graph data structure with cPickle; I'm seeing superlinear performance > > (plotting it seems to suggest n^2 where n is the number of nodes of my > > graph), in the duration of the pickle.dump calls and I can't quite > > figure out why. > > Try gc.disable() before loading the pickle, and gc.enable() after. > > > Is cPickle's behavior known to be O(n^2)? > > No, but the garbage collector's sometimes is.
Wow, that totally fixed it -- we went from 1200 seconds to 60 seconds. I'm somewhat shocked -- is the occasionally-abysmal behavior of the GC documented anywhere? ...Eric -- http://mail.python.org/mailman/listinfo/python-list