On Wed, Jan 7, 2009 at 8:44 AM, Chris Anderson <[email protected]> wrote: > On Wed, Jan 7, 2009 at 12:26 AM, George Palmer <[email protected]> > wrote: >> Assuming I've used the view once so the index is built, am I just >> paying the cost of looking up a userid as a key on an indexed data >> set, so performance is really comparable to selecting 25 from 250,000 >> in a RDMS (with an index on a table)? > > Correct, once the index is built, the view lookups are fast. > >> >> Also how does performance scale as the number of items in a view >> raises? What if that was 100million items in the view that we had to >> filter by userid? Obviously we're looking to use a big data set here! >> > > The lookup cost should be O(log N), due to CouchDB's Btree indexes. So > it should not slow down appreciable as the view index size grows. >
Another note that this is a large base log. Btrees are specifically designed to be shallow but very wide. > -- > Chris Anderson > http://jchris.mfdz.com >
