RE: Wow! Must-read paper.

Date view Thread view Subject view Author view

From: Edward Jung (edward@jungco.com)
Date: Mon Jun 05 2000 - 18:35:54 PDT


Although I hate to admit it, I co-wrote a patent on this idea at Microsoft
for COM monikers in 1993 (see US5581760 and several related patents). Not
nearly as well written, but the same idea.

The basic idea is that you can compose one or more bag of bits into a name.
Each bag is evaluated in the context of the name, which is the way it is
like a closure. The bag may be a query or a "classic" name, allowing you to
unify seach and navigation. It also allows you to unify several different
namespace handlers, e.g. an application document model, property namespace,
and a network naming system, and allows external programs to decompose and
recompose the name bits without understanding the naming conventions inside.
There is a compositional calculus of monikers.

Although monikers are typically associated with containers to resolve them,
they don't have to -- they can point to any resolver in any (non
hierarchical) context. The Cairo operating system (the killed
object-oriented OS from MS) used this mechanism to name and query everything
in a distributed network, also for distributed link management and repair,
and also had some features to handle various kinds of transitive closures
for bit copies and disconnected operation. The cool thing is that you could
build a three pane query system -- one pane for scope, another pane for
query, and a third pane to browse results -- and all the browser had to do
was to compose the names for the three enclosed panes (which could be
embedded objects) and presto! it would work. You could then add any pane to
the system you wanted.

I will admit that my work on this was made massively more complex by the COM
program managers. The original moniker interface had only five methods and
was simple. The current moniker interfaces are obfuscated to the point where
it's difficult to see how this all works. I also originally did not want an
independent composite moniker, but it made it easier for developers to
understand than generic composition.

Edward

-----Original Message-----
From: Jeff Bone [mailto:jbone@jump.net]
Sent: Tuesday, May 30, 2000 11:48 AM
To: fork@xent.com
Subject: Wow! Must-read paper.

I have just finished reading a paper that has warped my fragile little
mind. This might be the most important paper I've read since the
Lifestreams papers 5+ yrs ago: death 2 hierarchical naming, namespace
unification, dirs as closure, searching v. navigating, queries v.
containers, unstructured v. structured data, information retrieval,
etc... hits all my hot buttons. It's at.

    http://devlinux.com/projects/reiserfs/whitepaper.html

Particular concepts, though in some cases these points were made first /
better in other research / ongoing work, cf. Synopses, etc.

    * Pathnames in hierarchical namespaces are closures
    * Directories and queries are equivalent
    * Queries are names
    * Two naming primitives: ordering and grouping
    * Unification of filtering, fileing
    * Elimination of the notion of containers as first-class
    * Unification of navigation and searching
    * Containers / directories are intersection of names and attributes

The first part is a plainly written survey of the problem space I've
been edging about for years -wrt- the things mentioned above. It's my
intuitive belief that there is, out there somewhere in Platonic
Formland, ;-) a sort of Grand Unified Abstraction for simplifying many
of the problems inherent in information retrieval and use. Things like
UNIX, Plan 9, Linda, Lifestreams, and so forth all hint at and validate
this. Curmudgeons will say that there's no such Grand Unified Beast,
that one size does not fit all with respect to the actual problems these
different systems solve; different problems demand different
techniques. I'm not so sure: look at the power gained by namespace
unification in i.e. Plan 9 v. UNIX.

At any rate, here's somebody else that's thinking about this. This
particular somebody, Hans Reiser, has a working Linux filesystem which
is the research basis for this work; and his work is being funded by,
among other things, MP3.com (find me songs about trains, perhaps?) and
Ecila, a European search engine company.

This is wild. Jeff Bob sez check it out!

;-)

jb


Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Mon Jun 05 2000 - 18:39:21 PDT