One of the rules of efficiency is to avoid search. Our previous example of using an iterator over the keys in a shelf is inefficient. To state that more strongly, use of search defines an inefficient application. We'll emphasize this.
Brute-force search is perhaps the worst possible way to work with data. Try to design indexes based on subsets or key mappings to improve performance.
To avoid searching, we need to create indexes that list the items users are most likely to want. This saves you reading through the entire shelf to find an item or subset of items. A shelf index can't reference Python objects, as that would change the granularity at which the objects are stored. An index will only list key values, a separate retrieval is done to get the object in question. This makes navigation among objects indirect but still much faster...
