Abstract — We measure the performance of the Zopfli compression algorithm and compare it with other implementations of deflate compression. We show that Zopfli has the highest compression density of all deflate compatible algorithms we compared, on four compression corpora we used for testing. Zopfli uses significantly more time in compression, but decompression speed of Zopfli-generated output is comparable with other algorithms.
Zopfli is a new deflate compatible compressor that was inspired by compression improvements developed originally for the lossless mode of WebP image compression. Being compatible with deflate makes Zopfli compatible with zlib and gzip. Most internet browsers support deflate decompression, and it has a wide range of other applications. This means that Zopfli-compatible decompression is readily widely available.
In this study we compare the compression density of the Zopfli compressor with the compression density of zlib [1], the most common deflate implementation used today, as well as two lesser known deflate implementations, 7-zip [2] and kzip [3]. We use four data compression corpora [4–7] to measure the compression density. We also measured the compression and decompression speeds for one test corpus. In the conclusion we suggest a potentially important use case for the new Zopfli data compression algorithm.
Data compression works by eliminating statistical redundancy from the data. The redundancy can be, for example, in the form of some symbols occurring more often or sequences of symbols repeating.
There are many benefits to higher compression density. The smaller compressed size allows for storing more items in less space, faster data transmission, and lower web page load latencies. Furthermore, the smaller compressed size has additional benefits in mobile use, such as lower data transfer fees and reduced battery use. The higher data density is achieved byusing more exhaustive compression techniques, which make the compression a lot slower, but the decompression speed is not affected.
We chose several sets of files (corpora) for running the compressors:
For running the benchmarks, we used an Ubuntu-derivative Linux operating system with kernel 3.2.5 (x86_64) on Dell Precision T3500 Intel Xeon CPU X5650 running at 2.67 GHz. The versions of the various software used in experiments are Zopfli (https://code.google.com/p/zopfli/source/browse/ revision acc035299f8d), gzip 1.4, 7-Zip (A) [6,4] 9.20, and kzip (release 20091108). The compiler we used is gcc version 4.6.3. We run Zopfli and kzip with default arguments, gzip with -9, and 7-zip with -mm=Deflate -mx=9.
Compression results (Table 1) indicate that Zopfli produces the most dense output, but is slowest (Table 2) of all the algorithms we tested. Uncompression time (Table 3) is unaffected by the selection of the compression algorithm.
| Benchmark | Corpus size | gzip -9 | 7-zip | kzip | Zopfli |
|---|---|---|---|---|---|
| Alexa-top-10k | 693'108'837 | 128'498'665 | 125'599'259 | 125'163'521 | 123'755'118 |
| Calgary | 3'141'622 | 1'017'624 | 980'674 | 978'993 | 974'579 |
| Canterbury | 2'818'976 | 730'732 | 675'163 | 674'321 | 669'933 |
| enwik8 | 100'000'000 | 36'445'248 | 35'102'976 | 35'025'767 | 34'995'756 |
| Compression algorithm | Compression time |
|---|---|
| gzip -9 | 5.60 s |
| 7-zip -mm=Deflate -mx=9 | 128 s |
| kzip | 336 s |
| Zopfli | 454 s |
| Compression algorithm | Uncompress time for "gzip -d" of enwik8 |
| gzip -9 | 934 ms |
| 7-zip -mm=Deflate | -mx=9 949 ms |
| kzip | 937 ms |
| Zopfli | 926 ms |
Zopfli gives smaller deflate-compatible output size than gzip (3.7–8.3% smaller), 7-zip, and kzip, with more CPU time used at compression phase. This makes Zopfli ideal for uses where the cost of CPU is small in relation to the output size. Such use could include denser compression of static content for making websites faster.
Zopfli and gzip compress to gzip format, whereas kzip and 7-zip compress to zip format. This may alter the sizes slightly as the container format has slightly different overhead. For enwik8, the header overhead difference is below 0.0001% in relation to the output size.
7-Zip can operate with the deflate format, but it can read and write several other archive formats, and achieve higher compression ratios. In this study we only measured deflate-compatible compression.
We could achieve faster results with gzip and other algorithms by specifying lower compression density options. In this study we are interested on finding the smallest possiblecompressed size, and because of this we have only run every algorithm with maximum compression options. Zopfli also can run even longer to achieve slightly higher compression density, but we chose to run it with default settings.
In the light of the results we presented, we recommend Zopfli for compression of static content and other content where data transfer or storage costs are more significant than the increase in CPU time. To our knowledge Zopfli typically produces the highest compression density of any deflate-compatible algorithm.
Zopfli is opensourced at https://code.google.com/p/zopfli/. We invite everyone to try it out, and hope that it will find many practical uses.