[tex-live] TL-CD compression

Reinhard Kotucha reinhard.kotucha at web.de
Mon Jun 4 02:03:59 CEST 2007

>>>>> "Giuseppe" == Giuseppe Ghibò <ghibo at mandriva.com> writes:

  > If you give me the URL of the uncompressed .tar (but it shouldn't
  > contain any .zip file, as those can't be compressed further) I
  > will post some benchmark and compress ratio comparison using lzma.

The files are too large to download them from my machine.  But I did
some tests myself.

du -sc TL2007-*
598732  TL2007-t7z
655660  TL2007-tbz
695896  TL2007-tgz
728072  TL2007-zip

The gain of using 7zip is about 20% in file size.  That's about 130~MB.

Uncompressing the archive directory takes about a minute for the zip
files, a little bit more than 2 minutes for the tar.7z files and about
3 minutes for the tar.bz2 files.  The files had been read from a hard
disk.  My machine is as an amd64 2.2 MHz dual core.  I assume that
even on a 300 MHz machine CPU time doesn't matter.

It seems that 7zip is the winner.

But there is another problem: 7zip is probably not available on all
platforms.  Suppose we put binaries for all platforms on the cd and
let config.guess find out which one to use, we would lose an enormous
amount of disk space because the size of the binary (7zr) is 800 kB.
The size of the bzip2 binary is 33 kB.  bzip2 is quite popular today
and we can expect that it is installed at least on all Linux systems.

Though 7zip is available as a Gentoo Linux package, I downloaded the
sources.  There are makefiles for many platforms but I'm not sure that
all platforms which are supported by TL are supported by 7zip.
Another thing which probably will cause headaches is that it is
dynamically linked against the C++ libraries.
Linking them statically would blow up the binaries even more.  


Reinhard Kotucha			              Phone: +49-511-4592165
Marschnerstr. 25
D-30167 Hannover	                      mailto:reinhard.kotucha at web.de
Microsoft isn't the answer. Microsoft is the question, and the answer is NO.

More information about the tex-live mailing list