Thanks to all for your help. My graphs are huge (and contain much more information than the list of nodes and their edges... I am working with protein-protein interaction networks) so I have decided to have a persistent object on the server. I first had to make my objects "pickable" (which was not obvious, since they also contain database connections) but I think now I am on the right way. The only problem I see is that the web interface will be slower, since it has to "load graph-process it -dump graph" after each command, but I guess that is the price I'll have to pay.
Thanks a lot again! Ramon -- http://mail.python.org/mailman/listinfo/python-list