[CI] About uncorrelated cost model

[Samuel Madden]

I agree that a unified model might be good -- I just don't think it  
makes sense to use c_per_u in a cost model for uncorrelated attributes.

On Jun 24, 2008, at 7:53 PM, Hideaki Kimura wrote:

> I understand the cost model of uncorrelated case is correct, but it  
> might give reviewers a wrong impression that we are claiming that B 
> +Tree does something stupid in the case and our CT doesn't.
>
> As the CT base cost model shifts to sequential scan if c_per_u is  
> high (which is almost accurate in figure 3), I think we don't need a  
> different cost model for uncorrelated case.
> -- 
> Hideaki Kimura <hkimura at cs.brown.edu>
> _______________________________________________
> CI mailing list
> CI at list.cs.brown.edu
> http://list.cs.brown.edu/mailman/listinfo/ci