[gdal-dev] gdal_rasterize increase memory usage

Nicolas Cadieux nicolas.cadieux at archeotec.ca
Thu Jul 16 07:36:25 PDT 2015


I had the same problem with gdal_warp.  I recompiled the code using the latest ms visual studio community and the buffer was increased from 15 Mb to 1.5 GB.   It's was a first for me because I  usually work in python. It's easy to do.

That fixed all the issue.  I am told that buffer size is arbitrary and what may be good for one person may cause problems to another if, for example, a person launches many instances of the same program in parallel.

Best to recompile if the option is there. There is a page on the gdal web site explaining how to do it.  

Good luck

Nicolas Cadieux M.Sc.
Les Entreprises Archéotec inc. 
8548, rue Saint-Denis Montréal H2P 2H2
Téléphone: 514.381.5112  Fax: 514.381.4995

On Jul 16, 2015 05:15, jramm <jamessramm at gmail.com> wrote:
> gdal_rasterize is limited to use just 10MB of memory (line ~640 of 
> gdalrasterize.cpp). 
> Is there anyway to change this (without having to recompile?) 
> I'm noticing that changing the output data format from Byte to Int16 or even 
> Int32 drastically reduces performance. This must be because the strict 
> memory limits imposed by gdal_rasterize means that the amount of data it 
> reads in one go is exponentially reduce each time I increase the integer 
> size. Thus it has to loop through the polygon set many more times. 
> I'm seeing drastic slow down by just changing from Byte to Int16 when using 
> gdal_rasterize (8-10 times slower) 
> Given that even the cheapest desktop has at least 1GB of RAM, isn't the 
> scanline buffer size over conservative? 
> -- 
> View this message in context: http://osgeo-org.1560.x6.nabble.com/gdal-rasterize-increase-memory-usage-tp5215862.html 
> Sent from the GDAL - Dev mailing list archive at Nabble.com. 
> _______________________________________________ 
> gdal-dev mailing list 
> gdal-dev at lists.osgeo.org 
> http://lists.osgeo.org/mailman/listinfo/gdal-dev 

More information about the gdal-dev mailing list