3 Replies Latest reply on Mar 31, 2014 2:29 PM by B Hall

    Any Bulk R-Tree loading enhancements being coded for or beyond?

    B Hall

      I ask because as a routine measure, we create lots of tables/mv's with 10-100s of millions of rows of 2D point data (rollups - no partitioning). As part of that we of course need to create/refresh spatial indexes to make use of locator/spatial functions.


      The problem comes in that we do this often (daily in many cases). Even running parallel, we normally see the r-tree index creation time take about 60x-70x longer than a similar b-tree x/y index. Ouch.


      So the question - is anyone in Oracle looking at coding a bulk-load optimized create index option? Something like a Hilbert curve with random sampling using the forthcoming mapreduce db functionality should help greatly over the current implementation, at least based on results others are showing.


      Or is the plan to just drop spatial indexes altogether for simple points, and rather just load the spatial column x/y/(z) data into memory and search there instead of using a indexed approach?