[elephant-devel] Re: traversing btree using multiple indices

[Ian Eslick]

> IE> As for Sean's request about doing an efficient intersection, as  
> Alex
> IE> was explaining there are only two ways to do this efficiently:
>
> no, these are two ways to do it inefficiently, there is only one way  
> to do
> it efficiently -- via combined index :)

True, but his request was efficiency without an extra index - and  
those are the two ways of doing it faster than linear!  :)

> IE> The new unstable branch removes the MOP overhead so if your set  
> sizes
> IE> are in the low 10's of thousands this should take less than a  
> second.
>
> if it is slow or not depends on a use case -- one thing is interactive
> application where queries are made in response to user action and few
> seconds of delay are OK.
> another thing is a web site, where one slow query affects all users  
> using
> this web site concurrently.

True, I don't have a good sense of how all these things play together  
yet in a real world, multi-user setting.  Disk space is cheap, so you  
might as well use indices to save disk bandwidth!

Ian