[nycbug-talk] transparent clustering

[pete wright]

On 3/8/06, Josh McCormack <joshmccormack at travelersdiary.com> wrote:
> I'm listening to the Interview with Matthew Dillon from DragonFly BSD
> on http://bsdtalk.blogspot.com/
> With my limited understanding, I'm intrigued by the idea of
> transparent clustering.
> I've often thought that a way of putting machines together to share
> work (like if you had a laptop and you could have it work along with a
> desktop, without tons of special configurations) would be incredible.
>
> This sounds somewhat like that... am I completely off?

hmm...this is a kinda a really broad topic.  You can achieve
"transparent clustering" via something simple like a load balancer/SLB
(good for web/application server farms for example).  If you are doing
computation, you could approach it from a cue based perspective. 
That's sorta transparent as the end user submits a job and it goes to
the farm for processing.  Or you could go the openMosix/SGI-NUMA route
where you have a cluster of machines, sharing memory and working on
data set's together.

I belive Matt's looking at it from the latter perspective, which is
great but does have it's limitations.  More often than not, you need
to make sure your program is going to behave well in such an
environment (i.e. trying to render a Maya file in such an environment
will not work as it does not play nice with shared memory).  In the
end of the day I guess it depends on what you are trying to
accomplish.

-pete



--
~~o0OO0o~~
Pete Wright
www.nycbug.org
NYC's *BSD User Group