You can fully randomized uuids by changing from 'sequential' to 'random' but, yes, the shared prefix is a compromise between random uuids and values that play nicely with a b+tree.
B. On 30 June 2011 20:04, Jens Alfke <j...@mooseyard.com> wrote: > > On Jun 30, 2011, at 12:35 AM, Benoit Chesneau wrote: > >> Other way is to get a bunch of uuids from couch and using them in your >> PUT, so you don't have to generate yourself uuids. (and you have best >> of both world). > > Yes, but it’s much simpler (and faster) to call CFUUIDCreate() rather than > write all the code to manage a pool of available IDs, call the server on > demand to fetch more, block if necessary until the server replies, handle > HTTP errors… > > > > On Jun 29, 2011, at 6:08 PM, Dale Harvey wrote: > >> CouchDB generates uuid's that attempts to minimise the amount of rebalancing >> that needs to happen on insert > > That’s funny — I had noticed yesterday that the doc IDs have long prefixes in > common, and felt that this was possibly bad for performance because it slows > down string comparisons (more bytes to scan before finding a difference.) But > of course that’s trivial compared to the overhead you refer to of fetching > more b-tree pages from disk on insert. This must indicate that I come from a > hash-table background, not a b-tree one :) > > —Jens
