[NCLUG] Judy is for real

[Alan Silverstein]

Greetings.  Matt Taggart directed me to your recent discussion about
Judy and the LGPL.  It appears the simple way to post a reply is to
email to this address, so here goes.

Judy is for real in several ways.  First, the performance in both time
and space is as advertised.  While I worked on the main patent
applications I read over 20 related academic papers and 5 textbook
sections, and no one has anything like Judy.  It combines the speed of a
wide digital tree (wide and shallow, minimum CPU cache misses, and
simple, no balancing needed) with the space of a more efficient sort of
tree (through various compressions), all hidden below a remarkably
simple interface with no initialization, tuning, or configuration needed
or even possible, for all kinds of index sets (sequential, clustered,
periodic, random), with incredible scalability to the peta-range.

Second, HP really did LGPL the sources less than a month ago, see
http://sourceforge.net/projects/judy.  We did this because the project
was cancelled after about 2.5 years when three of the four engineers on
it took early retirement.

However, the Judy Shop Manual which illustrates the data structures,
etc. is not yet open-sourced, so to understand Judy (not just use it),
you'll have to make do with the comments in the source code, plus some
documents at www.hp.com/go/judy (while they last).

I've worked on Judy since before there was an official HP project.  I'm
not the source maintainer, though, that's Doug Baskins,
doug at sourcejudy.com.  Feel free to email me with questions, consultation
requests, etc, at least so long as *I* can remain at HP (might be
short-term).  Or email Doug.  I can probably mail you sample programs,
other documents, etc.

Thanks,
Alan Silverstein, ajs at fc.hp.com