On Wed, Jan 7, 2009 at 2:28 PM, Brett Henderson <[email protected]> wrote:
> Karl Newman wrote: > >> Heh, I was thinking along those lines, too, but I thought I'd wait until >> your format stabilizes. It would be quite handy to have a random-access data >> source for Osmosis without requiring a database installation. >> > I'm not sure if you realise this but there already is one of those. I > can't reach the wiki at the moment but the tasks in question are: > --write-customdb > --read-customdb > > I forget the details of the on-disk format but it's built around the > osmosis store classes so uses osmosis specific object serialisation. > > The --read-customdb task doesn't actually stream data to downstream tasks, > it provides the data in the form of a "dataset" object. This can be > consumed by tasks such as --dataset-dump which will stream from the dataset, > but it can also be consumed by any task wishing to access data randomly. > > Performance isn't great for large datasets which is why I created the > pgsql-simple schema. I've never been able to get an on-disk format to scale > nearly as well as a real database. > > I also wrote a set of tasks based on one the Berkeley DB Java Edition but > deleted it because performance was even worse. > > Brett > Actually yes, I was aware of that one, but I recall that you said the performance was not good and I thought you had removed that one for performance reasons (I didn't pay close attention; I guess it was the BDB that you removed instead). Karl
_______________________________________________ dev mailing list [email protected] http://lists.openstreetmap.org/listinfo/dev
