Subject: Re: FSBs and client-server
From: Frank Hecker <>
Date: Wed, 24 May 2000 23:28:25 -0400

Ian Lance Taylor wrote:
> It occurs to me that what is
> happening is that we are returning to a client-server architecture.
> Client-server was itself a return to the mainframe architecture.  In
> fact, computers move in a wave from centralization to decentralization
> and back.

This is also reminiscent of the "cycle of reincarnation" identified by
Myer and Sutherland with respect to doing graphics processing inboard or

> Now we're moving back to centralization, as people use browsers to
> connect to services provided from centralized systems.

But you can also see the beginnings of a move back to decentralization
as well, as people run more and more client software that leverages
Internet services but falls outside the standard browser paradigm. This
trend started with things like instant messaging clients (especially
souped-up ones like ICQ) and media players (RealAudio, WinAmp, etc.);
it's continuing now with clients that also act as servers, like
Gnutella, and with technologies like XUL that allow people to easily
build web-enabled clients that don't look anything like conventional

Note that the most recent examples of these programs/technologies (the
Gnutella clones and Mozilla/XUL) are libre software. In particular I
think XUL and XUL follow-ons will lead to a lot of libre software on the
desktop,for the following reasons:

* The technologies are already implemented as libre software, so they
can evolve freely in response to hackers' interests and needs.

* XUL files for existing applications (e.g., Mozilla and Netscape 6) are
not obfuscated but are easily visible (like downloaded HTML), so hackers
have a ready source of software to use as a base to start with.

* XUL hacking is open to anyone with a reasonable knowledge of HTML and
JavaScript; the potential XUL hacker population is thus several times
that of the population of developers familar with C/C++/GTK/etc.

> Libre software is most meaningful during a decentralization phase.
> When control over which programs are run resides with the end user,
> libre software is an enabling force.  When control over which programs
> are run resides in a central system, libre software is mainly of
> interest to the central system administrators.

Of direct interest, yes. However even in a centralization phase users
have an indirect interest in libre software, because libre clients act
as a check to the possibility of being locked into particular servers
due to the use of proprietary protocols. (And vice versa, of course,
which is of interest in the decentralization phase.)

> What's interesting to me right now is to speculate about the wave of
> decentralization which should follow the current wave of
> centralization.  What will it look like?

As I noted above, I think we are seeing the first glimpses of this now.

> Decentralization will consist of a small but powerful system which can
> tap into these network services and present them in clever and useful
> ways.  Libre software will flower even more as people trade around
> improved access and visualization engines.

Agreed: what else is Gnutella but a radical new way to do net searching
and information access? Andy Oram did an interesting article recently on
this subject:

He also discusses Freenet, which is (again) a libre application.

Frank Hecker            work:        home: