[geophile]

> using a Hilbert Curve for sharding doesn't seem like the best approach. You can partition by anything you like, it doesn't have to be arbitrary points along your index. Using 1-dimension to shard 2D data isn't optimal.

If you want to shard by proximity (items close in space are likely to be in the same shard, then the transform to 1d is the way to go, why wouldn't it be? What is your definition of "optimal"?

Sharding by proximity or by something else depends on the relative frequency of queries by location or something else. If you shard by location, then a query to one location goes (ususally) to one shard. That should scale better. Otherwise, each location-based query goes to each shard.