Comment 6 for bug 183156

Note also that:
(for f in launchpad/.bzr/repository/indices/*; do bzr dump-btree --raw $f; done) > raw_content.txt
Ends up with about 140MB of raw content. This doesn't account for the fact that:

1) Pretty much every key will be in there 2 times, if not significantly more than that. (each child will reference its parent, each file_key has a revision_id that is duplicated in the revision graph, and a file_id that is repeated as well.)
2) In memory, we will be using tuples of strings, which adds a fair amount of overhead. Specifically something like, 24 bytes of memory for a string, + 24 bytes of memory for a tuple (12 for the tuple fields, 12 for the gc fields) + 4 or 8 for the tuple pointers .