Bruce McFarling <agila61@[EMAIL PROTECTED]
> writes:
>On Mar 25, 11:43 am, an...@[EMAIL PROTECTED]
(Anton Ertl)
>wrote:
>> Grid computing is something very different from what you seem to
>> imagine. It has nothing to do with laying out processing elements in
>> a grid on a CPU chip; instead, it's a new marketing name for
>> supercomputer centers (there are also some new ideas).
>
>The argument is that getting a distributed processing paradigm to
>scale up to more capable processing elements is something that is
>likely to face fewer binding constraints than either getting a
>distributed processing paradigm to scale down or getting a programming
>model for a few processors to scale up.
>
>If you had a 5x8 grid of Seaforth chips and were successfully
>programming for 1,000 distinct processing elements as limited in
>capability as the Seaforth elements, there is every reason to believe
>that the programming model would scale up to massively superparallel
>with fewer hick-ups than a programming model developed for one through
>four processing elements.
I am not an expert in Grid Computing, but I have heard a number of
talks about it, and my impression is that the problems that they are
working on are very different from what you would ever encounter with
the kind of machine you are thinking of; and vice versa, in a SeaForth
chip you encounter issues that you will not encounter in a
supercomputing node.
Concerning 1000 processing elements, 85% of the top 500 supercomputers
have more than that <http://www.top500.org/stats/list/30/procclass>,
so there is no need to build a new machine if one wants to test that.
I also believe that it is easier to develop, say, some wheather
prediction code by starting on a PC than by starting on a 5x8 grid of
SeaForth chips; a big reason for that is that the local RAM of the
SeaForth processors is so tiny that you would have to organize such a
program very differently than from you would normally do for a
supercomputer.
I think the role of the SeaForth chips is not as a kind of small
supercomputer, but to do things that have been done with custom
hardware or field-programmable hardware before, with the advantages of
the SeaForth chips being:
* over custom hardware: more flexibility, reduced mask creation cost
(especially important for applications without huge quantities).
* over field-programmable hardware: possibly higher performance
(depending on the application). The different programming model may
be more useful for some applications (I guess there is a reason why
the FPGA companies support processor cores for their chips).
- anton
--
M. Anton Ertl http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
New standard: http://www.forth200x.org/forth200x.html
EuroForth 2008:
http://www.complang.tuwien.ac.at/anton/euroforth/ef08.html
|
