[
Lists Home |
Date Index |
Thread Index
]
On Friday 28 December 2001 12:11 pm, Champion, Mike wrote:
> Right, but maybe it's like saying we've had FTP and SGML for years that
> are roughly akin to HTTP and XML, so HTTP+XML=SOAP is no big deal.
I don't think so. Part of my background is in distributed systems
(hypermedia/workgroup systems)... so I'm speaking somewhat from experience.
There are some problems (Sean's XGrid) where the Linda-like model works well.
These tend to be compute-intensive, and/or parallel problems... and they
require a degree of skill to develop program for (certainly a mentality). One
way or another, Ruple adds little to traditional tuples-spaces (noting that
tuple-spaces can vastly simplify *some* applications as compared to PVM et
al.).
> much brandy in the Christmas fruitcake, but I can envision the possibility
> that XML and NXDB's will do for Linda what HTTP/HTML did for hypertext
> theory.
This is where we disagree. We've had SGML/XML repositories for years (about 6
or so), and while NXDB's are a key component for some things *they are not
themselves a killer application*. My questions are "where's the killer app?"
and "what's the ROI?".
There are a lot of applications that probably *could* benefit from a
Linda-esque model because synchronicity isn't needed, but these do not need
Ruple spaces per se.
> It's true that loosely coupled distributed systems have traditionally been
> hard to work with. I don't have enough knowledge of the details, but
> consider the SOAP analogy: CORBA and DCOM systems are hard to configure,
> make secure but useable, and a real pain to make them play nicely with one
> another. By combining the HTTP infrastructure with XML's platform
> neutrality and universal data representation capabilities, SOAP makes a lot
> of the traditional problems just go away. Similarly, the
> Linda/JavaSpaces/TSpaces folks have been trying to figure out how to make
> it possible for ordinary programmers to work with loosely coupled
> distributed systems for years. My hope is that by combining the Linda-esque
> loosely coupled distributed systems architecture with the XML physical
> infrastructure, Ruple, JXTA-spaces, etc. will achieve a similar
> transcendence.
There is some truth to the ease-of-use of SOAP vs CORBA. That has less to do
with HTTP and XML that it does with language neutrality, and simplicity in
the model (SOAP has no ORB, for example). The thing here though is that
making it more approachable doesn't take away the fundamental complexity of
the problem.
My wakeup call on this was in doing some work in the late 80's where we had a
system that worked wonderfully on a single machine, or local LAN, but once we
started bouncing packets to the US.. all hell broke loose. That system
was much like SOAP (used S-expressions and straight sockets though). I
realized then that it's *hard* to do transparent distributed programming
because the modes of performance, and modes of failure are so different.
I'm not sure how many people are using SOAP, and how many of those realize
that you have to design distributed applications somewhat differently from
monolithic applications.
> unpredictable, let's design around that fact." My highly non-expert guess
> is that while humans can deal with 404 errors just fine, that's going to
> throw the two-way XML web for a much bigger loop; the Linda-esque approach
> of connecting processes that are disconnected in space and/or time via
> "spaces" seems like a more solid foundation on which to build.
For some applications, I agree that tuple-spaces could be useful. Whether you
need/want Ruple for that is a different matter.
|
