WOFF 2.0 Compression Improvement Breakdown (preprocessing and LZMA)

As the early compression improvement results for the major font
collections are all in (thank you everyone!), we had an opportunity to
review the results in the last Web Fonts Working Group meeting.

The results (esp. average, median) are largely consistent (within a
few percent) across all of the collections, which is a great thing.
In general we are seeing 20-26% compression improvements for TrueType,
and 12-13% for CFF.  As the compression difference between the two
(TrueType and CFF) largely boils down to the extra preprocessing for
TrueType files, we discussed this further in the working group
meeting.

In the original proposal, Raph summarized the impact of the
preprocessing step and included graphs (see the following document):
http://wiki.font-compression-reference.googlecode.com/git/img/WOFFUltraCondensed.pdf

As a follow on to the discussion this week, I gathered the raw data
(compression improvement breakdown) for the Google Fonts collection:

WOFF 2.0 Compression Improvement Breakdown
https://docs.google.com/spreadsheet/ccc?key=0AvcH1ZzSrGMGdE9wYUtialItdU5hbmx4WVN4b3hHcWc#gid=0

And a summary of the breakdown:

          Prep   LZMA  Combined
Average 14.45%  11.96%  26.40%
Median  13.75%   8.76%  25.01%
Stdev    5.35%   9.27%   7.59%
Min      3.67%   1.63%  12.60%
Max     32.24%  53.11%  60.66%

While not represented in the Google Fonts collection, it's worth
noting that we have seen even larger compression improvement numbers
using LZMA for some fonts.  For example, Vlad and I were recently
looking at a font with a really large KERN table.  LZMA compression
was extremely effective in compressing it, resulting in a 90%+
compression improvement for this font over WOFF 1.0.

Received on Friday, 14 June 2013 22:30:03 UTC