Changes between Version 6 and Version 7 of PerformanceTesting
- Timestamp:
- Mar 10, 2009, 8:24:53 AM (15 years ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
PerformanceTesting
v6 v7 8 8 9 9 {{{ 10 11 Note that the state_shp layer is a big shapefile (>3,000,000 line features) and is drawn with a 1 pixel black line. 12 10 13 * Used the mapfile below out of the box with the extents below to draw a 500x500 square map with a square extent. I've only tested with one parallel request since I only have one CPU. 11 14 … … 19 22 * For the EPSG code performance problem, FrankW is planning to implement some caching directly in PROJ.4. More details to come, but on the PROJ.4 side, not a lot directly in MapServer. 20 23 * With really large shapefiles, some profiling revealed that a lot of time is passed in msGetBit called by either msShapeFileLayerNextShape and msFilterTreeSearch. We acknowledge that the bit mask is not the optimal solution for very large shapefile (>3M features). the only thing we plan to do right now is to pass from a 8bit mask to a 32 bit mask. 24 25 ==== Results ==== 26 27 || CGI request on state_shp EPSG code in map object and with PROJ.4 fix to cache EPSG file || 1 || 297 to 305 || 28 || CGI request on state_shp with the bitmask fix [http://trac.osgeo.org/mapserver/ticket/2930 ticket 2930] || 1 || 372 to 378 || 29 30 * It seems that the PROJ.4 fix removed most of the overhead of using epsg projection definition. 31 * ticket 2930 help a lot when reading big shapefiles. In our case, drawing small part of more than 3M features, we got around 20% more speed. 21 32 22 33 == Test machine ==