Also note that this is only true when reading over a 'dumb' transport. The smart server code reads things locally and streams the content over the wire, the indexes are not read directly.
A more involved option would be to set up a special case for .cix because we know we lose 'locality' there, and to only bump up the node cache size for cix.
So the trivial fix is: btree_index. py' btree_index. py 2009-08-13 19:56:26 +0000 btree_index. py 2009-08-14 18:55:59 +0000
=== modified file 'bzrlib/
--- bzrlib/
+++ bzrlib/
@@ -46,7 +46,7 @@
_PAGE_SIZE = 4096
# 4K per page: 4MB - 1000 entries
-_NODE_CACHE_SIZE = 1000
+_NODE_CACHE_SIZE = 100000
One could easily put this into a plugin.
Also note that this is only true when reading over a 'dumb' transport. The smart server code reads things locally and streams the content over the wire, the indexes are not read directly.
A more involved option would be to set up a special case for .cix because we know we lose 'locality' there, and to only bump up the node cache size for cix.