I'm currently trying to implement a distributed cost-distance algorithm, however I can't seem to invent my own and still make it fast. I've been googling around a bit but haven't found anything that would allow me to compute the cost-distance for a tiled raster in a "map-reduce" manner, e.g. that would run efficiently on Apache Spark. Does anybody have any good papers or reference implementations of this?
I'm working on GeoTrellis (https://github.com/geotrellis/geotrellis) In Scala, we have are using our own raster-processing library. I'm looking for papers essentially in this question.