News for hackers. Stuff that matters.
Re:One trillion edges: graph processing at Facebook scale [pdf] by jandrewrogers on Thursday August 13, 2015 @12:47AM (#10052302)
The primary advantage of Hilbert curves relative to Morton (Z) curves is that they are marginally more efficient at linearizing multidimensional point indexes e.g. storing on spinning disk.

However, the reason that Hilbert curves are relatively rare in real systems these days is that (1) the optimality of Hilbert curves is largely based on the assumption of indexing on sequential storage media, and (2) for the many other applications of structures describable by space-filling curves, a Hilbert description is often inferior in practice. For example, it is a suboptimal curve to use if you are implementing spatial analysis operators in a distributed system.

If you were designing a spatial indexing system from scratch today for modern applications, you would use neither Hilbert nor Morton/Z curves but they are adequate shortcuts for sufficiently narrow applications. Substantially more optimal curves also have a much more complex description.

Re:One trillion edges: graph processing at Facebook scale [pdf] by cbsmith on Thursday August 13, 2015 @01:40AM (#10052432)
What would you go with for spatial indexing today? R*-tree? k-d tree? BVH?