Diskussion:HSR Texas Geo Database Benchmark

Feel free to discusss 'HSR Texas Spatial Database Benchmark' issues here! Stefan 18:39, 22. Dez. 2009 (CET)

To do

• Standardise data and input process (circumvent or minimize floating point problem)
• Define the variables (@bbox and @point)

Observations

General:

• MS SQL Server representatives report that geography types are 20-30% slower than geometry types.
• About PostGIS 1.5 there was said, that geography types are somewhat faster (see http://blog.cleverelephant.ca/)
• Secondary index optimization makes performance about 50% faster (from: MS SQL Server discussion)
• The number of points of a query window is important for a secondary filter overhead.

Data and queries per se:

• These queries return quite large percent of the data overall (50%-100% for points and polygons)
• The Number of points for a query windows is very small.
• The query windows are "aligned bounding boxes" which may put in favor RTree (e.g. Postgres) or spatial grid (SQL Server) index, depending on a position of the window. If the window is aligned with the grid, performance can be much faster than when it's slightly off.

Discussion Points

• Should we take an average of multiple runs with different values (@point, @bbox)?
• Should we vary the queries' result sizes? (see observations)
• Should we vary the shape and alignment of the query window? (see observations)
• Should we run the benchmark with parallelized queries?

Review

• The limiting factor of Query 1 is mainly the performance of the harddisk.
• Query 1 is rather pointless without index on 'roadflg'. What should be proved with this query anyway?