>>>>> "Jim" == Jim Gettys <[EMAIL PROTECTED]> writes:

Jim> Therefore another piece of performance data we should gather on
Jim> client side fonts is how well glyphs compress,

Out of curiosity, I just compiled 100 dpi PKs of the ec1000 from (EC
version of Knuth's Computer Modern Roman at 10pt optical) at
magnifications i <= 200, i in Z+ to see how the byte scale in
proportion to the magnification factor.

This is roughly similar to generation bitmaps at point sizes from 10
to 2000 of a typical non-optically scaled font.

As I recall, the PK format uses a simple RLE compression (it was after
all designed for *much* slower boxen than we see today :), but I have
not looked it up to confirm....

The EC font has 256 (mostly) latin glyphs, and is a serif face.

I used modes.mf's mextscrn MF mode, which provies no corrections for
darkness, etc.  This is the default mode for 100 dpi in mktexpk(1),
and would typically be used to generate non-aa bitmaps for screen
viewing in X.  (For aa bitmaps, most dvi viewers use the 600 dpi
laserjet 4 mode and convert the resulting 600 dpi bitmap into a 100
dpi greymap for display.)

This is not a very scientific test, given the use of only one face,
but it does give some info.

The result is that PK's compression is subliniear on this face at mag
factors less than 14 and superlinear above mag factor 14.  It is very
close to exactly linear at mag factor 14.

I then converted the PK files to uncompressed GF files, and ran them
through 'gzip --name --best' and 'bzip2 -9'.  The gzip compressed GF
files showed sublinear growth at mag factors less than 200 and approx
linear growth at exactly mag factor 200.  Interestingly, bzip2 -9
transitions from sublinear to superliner between mag factor 138 and
139.  Of course, bzip2 has too high of a computational component for
use in this context; the update latency were it to be used to compress
glyph pixmaps for transfer from client to server would be excessive.
Even in the case of deflating the pixmaps, a faster deflate level
would be expected.

I'll put together a simple web page with the data and some graphs at
http://www.io.com/~cloos/pk/ if anyone is interested.  It should be up
by about 1800 UTC today.

-JimC


_______________________________________________
Fonts mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/fonts

Reply via email to