JSON compression in JavaScript

This page presents measurements of various compression techniques applied to 100000 bytes of real-world JSON data. Only forward transformations/compression was tested, no inverses and/or decompression. All tests were executed on Intel E6750 with 32bit Windows XP, Firefox 3.6.

Contestants

MTF - Move To Front transformation, implemented using a 256-byte array.
BWT - Burrows-Wheeler transformation, implemented using Array.prototype.sort() applied to an array of indexes.
LZW - Lempel-Ziv-Welch algorithm. Operates in two modes:
- variable output width (slower, default), using 8 to 16 bits per output number
- fixed output width (faster), using 16 bits per output number

Results

Single

Algorithm	Size (bytes)	Time (msec)
MTF	100000	330
BWT	100001	11011
LZW - variable output width	18274	174
LZW - fixed output width	22970	110

Remarks:

MTF and BWT are not compression algorithms; they are shown here just to demonstrate their performance.

Combined

Algorithm	Size (bytes)	Time (msec)
MTF + LZW	33000	540
BWT + LZW	9906	11159
BWT + MTF + LZW	7070	11577

Remarks:

It is currently not possible to run BWT after MTF, due to the way BWT is implemented (certain special byte, \0, must not be present in input data).

BWT + MTF + LZW using smaller chunks

Chunk size (bytes)	Size (bytes)	Time (msec)
5000	14956	5862
10000	11285	7110
20000	8869	8336
50000	7556	10209

TODO

Optimize BWT encoding implementation
Benchmark inverse transformations and decompression