[CI] Paper theme

[Samuel Madden]

Thought it would be good down our new pitch and agree on it.

We are claiming:

- A method for detecting "soft dependencies" (what we have been  
calling correlations) between attributes

- Given a collection of such dependencies, a physical organization of  
the database that includes:

a) A choice of a primary (clustered) attribute

b) A collection of secondary indices

- A technique for bucketing secondary indices to decrease their size  
without giving up performance dramatically

- A cost model that demonstrates we can predict which indices will  
benefit and what the benefit will be

- A collection of performance results that shows that this works well  
in many real cases

Other things?

Here are some things I am not sure about:

- Besides index compression, do we think what we are proposing is  
novel (e.g., are soft dependencies plus our cost model and search  
techniques new, even if just running on unclustered indices)?

- Are our index compression techniques new?

- How does this related to the Graefe paper the reviewers told us about?

Alex / Hideaki -- any insight as to the last question?

-Sam