![](https://secure.gravatar.com/avatar/a5de4cf6334caef556290f8bcd00f09a.jpg?s=120&d=mm&r=g)
On Feb 12, 2009, at 6:14 AM, Alexandre Fayolle wrote:
Le Wednesday 11 February 2009 19:49:59 Kelly F. Hickel, vous avez écrit :
Ahh, yes, now I remember. Psyco ONLY works with 32 bit python, whereas I MUST use 64 bit because the memory footprint for cvs2svn with our repo is just too large for 32 bit processes.
So, dead in the water again?
Hmm, just a quick question: are you sure your problem is CPU bound (as opposed to IO bound)? If it is not, you are looking at the wrong place to solve your performance issue, and you should rather look for a faster disk, faster network, etc. Also from what you say hereabove, maybe adding some RAM could help.
I'm answering this as we already talked about it in some private emails. He already figured out the problem is cpu bond, one of the cpus goes to 100% and stays almost the whole time this way for 6.5 days. His machine has 32 gb of RAM and the python conversion program uses good part of it that's why he can't run the thing on 32bit mode. His repo is only 6gb in size so I think maybe the problem is either a bad performing algo no cvs2svn or a bad cvs library (or whatever is used to access cvs). If it is the cvs library there is nothing no one can do (even psyco or pypy would not help). If the problem is a wrong pure python algo it should probably be changed (if a better one exists of course). Now why would a program reading a 6gb repo, even using a database file for intermediate data still uses tons of ram to do this conversion I don't know. -- Leonardo Santagada santagada at gmail.com