as tested on MySQL table for selecting locations by distance from a center point:
without indexes:
- n-vector is about 10% faster than lat/lon, due to less CPU intensive algorythm
with indexes:
- n-vector is about the same as lat/lon on small data (~30k rows), but gets worse on medium data (~200k rows) or large data
- on large data (~10M rows) it can be slower by order of magnitude (it may be due to 3-column index instead of 2-column, but my testing environment is crappy)
lon/lat in radians:
- when storing lon/lat in radians instead of degrees, selects are about 2% faster
- this is not enough to justify worse human readability of data
- lon/lat does not work arround poles and 180th meridian
- n-vector has false negatives (bug) and performance issues and needs further testing
it is recommended to use lon/lat in degrees for now