Re: multispectral file access

On Mon, 1 Sep 2014, [email protected] wrote:
>
> Yes, openimageio is very like vips, though they don't have (as far as
> I know) the fancy threading system or the range of operations. I don't
> know how speed compares. It'd be fun to add them to the vips
> benchmark:
>
> http://www.vips.ecs.soton.ac.uk/index.php?title=Speed_and_Memory_Use

It would be interesting to see how the GraphicsMagick results change 
for this benchmark with this environment variable setting:

   MAGICK_MMAP_READ=TRUE

This enables using mmap() when reading input files, which produces 
better-looking results (but likely higher RSS) for repeated 
benchmarks.

It used to be that GraphicsMagick used memory-mapping for input files 
if they would "fit" but I changed this default since GraphicsMagick is 
usually used for many files in one run and the many memory maps put 
unnecessary pressure on the VM system when the input data is 
continually changing.  Using memory-mapping decreases latency if the 
file is immediately accessed again but it clogs memory with file data. 
The filesystem cache is usually more intelligent than the VM system on 
modern systems.

To measure RSS under Linux, take a look at using 'time -v'.  Be sure 
to double check the reported size since in my experience the Linux 
kernel and 'time -v' don't agree on the meaning of rusage.ru_maxrss 
values and they are off by a factor of 4.

Bob
-- 
Bob Friesenhahn
[email protected], http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/

------------------------------------------------------------------------------
Slashdot TV.  
Video for Nerds.  Stuff that matters.
http://tv.slashdot.org/