[Zlib-devel] minigzip vs. my own zlib client

[Mikhail Teterin]

Although the e-mail below, apparently, never reached the mailing list (Mark is 
looking into the delivery problem), I have already solved the problem.

Using `9' for memLevel, as I do in the below call to deflateInit2, causes BOTH 
a slight increase in the output size, AND a very substantial performance hit. 
The comment in zlib.h says:

     The memLevel parameter specifies how much memory should be allocated
   for the internal compression state. memLevel=1 uses minimum memory but
   is slow and reduces compression ratio; memLevel=9 uses maximum memory
   for optimal speed. The default value is 8. See zconf.h for total memory
   usage as a function of windowBits and memLevel.

That is not true -- at least, not on my Opteron (FreeBSD/amd64). Using 8 is 
much faster AND more efficient, than 9... Looks like a bug...

	-mi

On Tuesday 30 May 2006 21:30, Mikhail Teterin wrote:
= Hello!
= 
= I'm writing my own zlib client, intended to compress files, which are almost 
= always cached in memory (just created). Knowing this, my program mmap-s them
= and calls deflate() with the very large buffers at once (512Mb is the 
= default).
= 
= It all works, but, for some reason, the minigzip-created archives are
= slightly (like 1.1%) smaller than those created by my program -- using the
= same -9 level. Both can be decompressed into the identical original, so
= there is no problem with correctness, just compressing efficiency...
= 
= Can anyone speculate, why this would be? I create the stream with:
= 
= 	deflateInit2(&stream.gz, level, Z_DEFLATED, 31, 9, Z_DEFAULT_STRATEGY);
= 
= What is minigzip doing differently? Thanks!
= 
= 	-mi