[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Gnu-arch-users] [FOSDEM substitute] "The Literature Shelf is not Litera
From: |
Thomas Lord |
Subject: |
[Gnu-arch-users] [FOSDEM substitute] "The Literature Shelf is not Literature" |
Date: |
Fri, 24 Feb 2006 17:10:51 -0800 |
Since I won't be at FOSDEM I'm not preparing a talk or slides. I am,
however, writing a brief series of short essays about the topic I
meant to speak about.
This is the second in that series of essays. The first was posted
earlier and is called "The King's English".
The Literature Shelf is Not Literature
W3C hyperlinks, as least as popularly used, never point to
documents. Instead, links point to hypothetical shelf space -- to
some location deep in the depths of DNS tables, routing tables,
network and server configurations. To refer to a document using
today's links, you must speak only of where you think that document
is likely to be stored.
As we saw in the first essay, that arrangement makes the privilege
of being a web author very expensive and pretty much eliminates the
possibility of creating long-lasting, high-quality archives of web
content.
Starting with the archival problem: Can we modify the web so that
documents can be linked to no matter where the document happens to
be?
* If There Were True Names
Pretend for a moment that every document ever or yet to be created
had, built-in, a globally unique name.
In that case, hyperlinks could be pointers to documents by name
rather than pointers by locations.
This would add an extra step to following a link. Instead of just
"read from http://abc.com" the first step of following a link would
be "find out where a copy of document://xyzzy is located," the
second step would be to read from that location, the third step
would be to make sure the result obtained is the document sought
after.
For relatively static documents, a system of true names suggests we
should have a global system of archives. An archive simply stores
and forwards as many documents as it can, indexing each by its
globally unique name (at least). Many archives might each contain
copies of a single document. When there are too many documents to
preserve, archives can apply criteria such as document popularity,
cross-links to a document, document newness, historic significance,
and sponsored storage.
For interactive documents, a system of true names suggests room to
define a set of host-independent services -- services defined as
text that can be run anywhere. Along with that: commodity service
hosting (of host-independent services) and on-the-fly routing and
scheduling to look up a "nearby", on-demand instance of a
host-independent service.
* You Can't Do It Unless You Do
Any competent, non-newbie software engineer should be able to tell
you that guaranteed, global True Names are a logical and physical
impossibility. Documents simply do not have natural, unique names.
Nothing can ever change this basic fact.
But then, a good software engineer will tell you how to implement
mostly-true True Names -- how to make a solution that's "pretty much
guaranteed to work most of the time." That's a topic for future
essays.
-t
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [Gnu-arch-users] [FOSDEM substitute] "The Literature Shelf is not Literature",
Thomas Lord <=