search

InitialDLab / Simba

Spatial In-Memory Big data Analytics
Apache License 2.0
120 stars 62 forks source link

Measuring distance for OSM dataset, paper #77

Open micvbang opened 7 years ago

micvbang commented 7 years ago

Hello Simba developers,

I am currently writing my master's thesis on parallelization of spatial queries. During this work, I am considering using Simba as one of the DBMSs to test.

In your paper, you show how Simba compares to other systems when performing spatial joins on the OSM dataset. As far as I can tell, Simba does not support geospatial queries, but only spatial queries. How have you ensured that the distance of your spatial join is consistent across systems? I.e. that you search for objects that are within exactly 1000m from each other.

Have you e.g. used a equirectangular map projection, allowing you to estimate distances using euclidean distance across all systems? From your paper, I cannot seem to figure out how this was done!

Thank you, Michael

dongx-psu commented 7 years ago

Simba is currently an experimental system constructed on Euclidean space. Thus, in our current implementation, we don't have any equirectangular map projection to calculate distance in different coordinating systems.

micvbang commented 7 years ago

Sorry, I forgot to reply to this issue.

Thank you for the quick response!

I think my solution will be to use the filter and refine approach. First, I will filter my data using Simba's CIRCLERANGE to compute an approximate distance between points using a a projection with units in menters, e.g. ESPG:3857, then refine using a more expensive and exact distance calculation, e.g. Haversine formula, implemented as a UDF. Perhaps using an existing implementation, such as spatial4j.