Maybe a little tricky to follow what’s going on without me wildly waving my arms around but I promised to put the slides up online (I used keynote converted to pdf, so you’re missing my notes, which were mostly swearing anyway) …
… to start with you have to understand 2 things:
- The talk was absolutely fascinating
- The joke about Pirates was hilarious
Beyond that, until I write it up properly by adding in the bits I realized I missed out or wasn’t clear enough about (thank you to the people who asked great questions at the end) these are the main take away points.
- If you’re hosting on a single box, you probably don’t need spatial indexing, by the time you need it you’ll have moved into a horizontal server setup.
- When you’ve moved onto a horizontal server setup, you’ll need some kind of indexing engine and you probably don’t need spatial indexing until you hit around 50 million documents, it’s just not really worth it, unless you’re doing something really complex.
- Hilbert Space Filling Curve is used for spatial searches. Until you start hitting limitations you can happily index on latitude and longitude. Using Hilbert’s you can collapse those two indexes down to one, either stopping your queries from taking too long, or freeing up an index for something else, like time, etc.
- Morton’s Curve is used for clustering, not spatial searching.
- Neither the Hilbert or Morton curve are optimal solutions, but they are good and easy enough.
- Morton’s Curve doesn’t magically do the clustering for you, you still need to build it into your clustering methods, there’s a billion other ways to do clustering. Morton has a certain elegance to it though
- Reverse geo-coding is a lot harder than you think it is
I’ll probably write up the Reverse geo-coding part first, as that’s the one that’s possibly more interesting than space filling curves, but who knows :)
Advertisement
Filed under: conferences, flickr, talks
