Re: Taos paper (fwd) (long)
- To: rdm@xxxxxxxxxxxxxxxxx
- Subject: Re: Taos paper (fwd) (long)
- From: Eugen Leitl <ui22204@xxxxxxxxxxxxxxxxxxxxxxx>
- Date: Thu, 17 Aug 1995 18:15:42 +0200 (MET DST)
- Cc: MISC
- In-Reply-To: <199508170020.UAA13396@magenta.com>
On Wed, 16 Aug 1995 rdm@tad.micro.umn.edu wrote:
> These people sure are modest, though I'm not sure if I really need to
> buy a holy grail.
Grant them some bit of grandesse:
these few guys really created an impressive design.
> More to the point, this paper doesn't give a hint of how to design
> properly for a taos system. [e.g. what kind of time penalty does
> message passing have? Where's the tradeoff between "the complete Open
In-node message passing is very fast. I don't know how they route
in the network, though. They are either using links or FIFOs with
other processors. Especially interesting is a distributed application,
dynamically allocating resources on a heterogenous system. This
has not been done before, ever.
I strongly recommend you to read Dick's article. I found the paper
I forwarded technically very weak as compared to his article.
> System" and "programs [that] can take advantage of special purpose
> hardware to run specific objects?"
They are using lots of small lightweight threads. Tool size is
typicall tiny. Implemenations language is TaosVM assembly, though
they planned a C++, etc. sometime. Taos port to a new processor
takes currently 6 man-months. Loadtime translation is purported
to be faster than a HD can provide data.
The point of Taos is to provide a consistant interface to
heterogenous software, lots of tools running on any system,
maspar msg-passing OOP paradim support and intelligent load
leveling.
Games and set-top boxes are application targets. Supercomputing?
I dunno.
> And, as with any sparkling new technology, I really want to know: how
> is it being disseminated into the educational system? Any industrial
> "literacy requirement" which isn't supported by educators is going to
> wither.
>
> [Hmm... I suppose some of these comments might apply to misc as well.]
This is the main point. Anything grand and new will perish if
not accepted by the market. I very much doubt Taos will be a mass
product. (And Dick reviewed it _two_ times. Anything he reviewed so
far sank as a lead baloon. Now it will be probably at least an osmium
baloon).
-- Eugene
P.S. I lost track of Taos' progress: it is not cheap and I don't have
a parallel box at home, yet. Lead baloon.
> --
> Raul
>