Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

With all due respect, 170M points is a trivial data model, it fits on a thumb drive. That is seconds worth of data in IoT; you can brute-force the implementation and look great. Several well-known companies do. It just doesn't scale to anything real.

The location analytics data models I need to support, e.g. mobile telecom, have trillions of polygons. Indexing millions of records per second continuously concurrent with sub-second queries is pretty normal for IoT, and easy to support with good computer science. (Location is almost never described as a point by primary sources but as polygons; these are converted to centroid approximations for platforms incapable of doing analysis on the source.)

While I do not know the details of your implementation, I do know that indexing complex geometries using a skiplist index won't achieve the necessary volume or velocity required for IoT-like applications. If you restrict yourself to point data, people have been building extremely fast, extremely large quad-trees for years; you just can't scale high-value analytics with them, which limits their utility.



Fair enough; though I'm intrigued by the mention of trillions of polygons. How complex are they? 10 points? 100? 1000? What benefit does the raw data give that centroids can't?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: