Size

The tools were measured on three criteria: file size, compression time, decompression time.

/bin/*

This test tries to compress the contents of my /bin directory. There are 97 executables to compress accumulating to 9Mb of data. As with all other tests: the lower the better.

7z	26.0%
ace	31.5%
arj	48.7%
jar	48.8%
lha	48.2%
rar	43.3%
sit	45.9%
tar.bz2	41.3%
tar.gz	47.8%
tar.lzo	51.0%
tar.Z	63.8%
zip	48.6%
zoo	63.9%

The big winners in this test seem to be 7z and ace. At 26% the 7z compacted archive is nearly half the size of most other archives. Ace also gets good compression but then there is a big gap before the third place. Bz2 showing here that it can beat Rar but they both are bunched with the other weaker compressors.

bitmap

The bitmap test file is a lovely picture of Will after being converted to a 24bit bmp. The file is 1.9Mb long and should compress very nicely.

7z	12.2%
ace	14.8%
arj	17.0%
jar	17.4%
lha	15.7%
rar	11.7%
sit	10.3%
tar.bz2	10.2%
tar.gz	16.7%
tar.lzo	21.6%
tar.Z	24.9%
zip	16.7%
zoo	30.3%

Sit and tar.bz2 are neck and neck here while ace and 7z are this time left behind.

html

The html test is just 3 html files (slashdot front page, google page searching for "compression", and my homepage. This would test how the tools deal with small amounts of data (in this case 148Kb).

7z	19.5%
ace	20.4%
arj	21.9%
jar	21.9%
lha	21.6%
rar	19.1%
sit	20.4%
tar.bz2	19.9%
tar.gz	21.3%
tar.lzo	24.5%
tar.Z	38.0%
zip	21.8%
zoo	43.3%

With the exception of tar.Z and zoo most compressors gave very similar ratios (with the usual suspects getting a percent or two better than others).

jpeg

Often archives contain compressed files and the archive is just used to stick them all together. In this test a set of Jpeg files (which are already compressed) were recompressed. Often archivers need to know when not to compress as compressing will give larger results than the original.

7z	97.0%
ace	97.3%
arj	97.6%
jar	97.7%
lha	97.5%
rar	97.7%
sit	97.4%
tar.bz2	97.8%
tar.gz	97.7%
tar.lzo	98.7%
tar.Z	122.0%
zip	97.7%
zoo	100.0%

In the tar.Z test is a perfect example of compressors not knowing when to just store uncompressed. The other compressors seem to have found little niches where to optimise the files but only by two to three percent.

pdf

Pdf files are a strange mixture of ascii and binary so a static ascii compression would not work well. Here the test was done on three large files from my research.

7z	58.7%
ace	60.3%
arj	62.0%
jar	62.9%
lha	62.2%
rar	60.5%
sit	61.7%
tar.bz2	61.0%
tar.gz	62.2%
tar.lzo	63.3%
tar.Z	88.4%
zip	62.1%
zoo	87.8%

I was expecting somewhat more interesting results from this one but the ratio distribution of the "good" compressors falls within 5%. As usual tar.Z and zoo perform poorly.

Linux

Here is a large test. The linux kernel takes up 88Mb. It is normally distributed in tar.bz2, but how do other compressors compare. There are over 4000 files most being C source. A good compressor would bundle all files of the same extension and compress them together.

7z	17.3%
ace	19.6%
arj	26.0%
jar	26.1%
lha	25.5%
rar	20.8%
sit	23.5%
tar.bz2	18.5%
tar.gz	23.2%
tar.lzo	26.8%
tar.Z	39.6%
zip	26.0%
zoo	41.6%

Well here you can see why they use bz2 to distribute the kernel. Even without being able to rearrange files it manages to be the second best. 7z wins the test.

Roms

A commonly compressed file is the rom images used in emulators. Here the roms being compressed are the X-MESS bios images. The files total 53Mb and are normally distributed compressed in zips.

7z	20.1%
ace	26.4%
arj	46.9%
jar	47.1%
lha	46.6%
rar	43.8%
sit	47.3%
tar.bz2	45.8%
tar.gz	46.7%
tar.lzo	50.6%
tar.Z	65.7%
zip	47.0%
zoo	60.6%

Here I had to redo the test several times because I could not understand how 7z managed to get such a good result. At 20% it is less then half the size of the zip. Ace also performs very well. The remaining tools do not come even close. There is a graph of all above tests.