The zdelta-2.0 Experimental Results




Experimental results comparing zdelta against vcdiff (version 2002), xdelta, and gzip.
Three different classes of experiments are presented here.
Experiments exploring potential delta compression applications. The experiments presented here were conducted on machines with the following configurations:
NOTE: Machine II and Machine III differ from the machines used to run the corresponding zdelta-1.0 experiments. For more information on the conducted experiments refer to the results section of [coming soon].

Experiments comparing zdelta-2.0 to other available delta compression tools:



Experiment 1


gcc size ratio MachineI compress MachineI decompress MachineII compress MachineII decompress MachineIII compress MachineIII decompress
uncompressed 27289 - - - - - - -
gzip 7563 3.61 31.1 17.9 8.52 6.11 33.8 22.8
xdelta 462 59.1 21.1 15.7 4.94 3.80 26.9 22.9
vcdiff 2002 287 94.9 31.9 16.3 11.3 5.83 42.2 20.6
zdelta (STDIO) 227 120 37.6 18.0 9.20 6.0 43.4 20.7
zdelta (direct) 227 120 29.6 10.0 5.66 2.41 32.7 11.6

Compressed sizes and running times for the gcc files (sizes in KB and times in seconds)


emacs size ratio MachineI compress MachineI decompress MachineII compress MachineII decompress MachineIII compress MachineIII decompress
uncompressed 27327 - - - - - - -
gzip 8577 3.19 35.8 22.5 10.2 7.73 40.6 29.6
xdelta 2132 12.8 30.1 21.2 10.2 7.73 40.6 29.6
vcdiff 2002 1819 15.0 34.8 20.8 11.8 7.50 44.6 26.3
zdelta (STDIO) 1419 19.3 49.7 22.7 12.7 7.51 57.7 25.5
zdelta (direct) 1419 19.3 39.4 12.5 8.22 3.08 44.5 13.9

Compressed sizes and running times for the emacs files (sizes in KB and times in seconds)



Experiment 2





Compressed size (in KB) versus file similarity





Compression time (in seconds) versus file similarity





Decompression time (in seconds) versus file similarity



Experiment 3





Compression time (in seconds) versus file size





Decompression time (in seconds) versus file size

Experiments exploring potential delta compression applications:



Experiment 4


Benchmark gzip trivial delta vcdiff 2002 zdelta
version size size ratio size ratio size ratio size ratio
Version2 (August 15)143MB 35.1MB4.08 12.6MB11.4 1.9MB74.5 1.46MB97.7
Version3 (September 2)143MB 35.0MB4.08 16.4MB8.69 2.8MB50.7 2.15MB64.2
Version4 (October 27)140MB 34.3MB4.09 21.0MB6.66 5.02MB27.9 3.86MB36.3

Compressed sizes for different versions of a web crawl consiting of 10000 HTML files. Reference Version1 crawled on August 13.


Benchmark gzip vcdiff 2002 zdelta
version size size ratio size ratio size ratio
Version2 (August 15)54.2MB 12.6MB4.30 1.78MB30.5 1.39MB39.0
Version3 (September 2)71.8MB 16.4MB4.37 2.69MB26.7 2.09MB34.3
Version4 (October 27)91.1MB 21.0MB4.33 4.33MB18.5 3.81MB23.9

Compressed sizes for different versions of a web crawl consiting of 10000 HTML files - compressing only non-identical pages. Reference Version1 crawled on August 13.

Experiment 5

BenchmarkSize(MB)Filescat+gzip ratiozdeltaratio
cbc 10.59 530 5.83 10.0
cbsnews 8.27 218 5.06 15.4
csmonitor 16.58 344 5.06 17.3
ebay 2.20 100 6.78 10.9
thomas 2.85 105 6.39 9.73
usatoday 8.23 344 6.26 9.17
emacs 38.86 1438 3.32 4.97
gcc 25.86 917 3.87 4.72

Compression ratios for for files with similar content and/or structure - that is pages from a given website, or source files for a specific program.

Experiment 6

The results will be made available soon.

Copyright © Dimitre Trendafilov 2002.
dtrend01@utopia.poly.edu