On Wed, Jul 28, 2004 at 10:45:53AM +0200, Reuti scratched on the wall:
> > So the question kind of brings us back to what Xgrid is, and what it
> > provides, vs. a more traditional cluster. Xgrid really is more of a
> > grid technology than a cluster technology, so the question is then
> > "what's the difference between a cluster and a grid?" which is a huge
> > discussion in and of itself.
>
> Also SUN is calling their cluster software "Grid Engine". Must there be a
> difference at all?
There is definitely a blurring line between the two concepts--
especially as low-cost cluster solutions that allow people to
reconfigure and rebuild clusters fairly easily. It could be argued
that these are not really formal, traditional clusters, however.
The term "cluster" in the most formal CS sense of things, is a *lot*
more than just a bunch of machines hooked together.
I think Sun's use of the word "grid" comes from two things. First,
what SGE (or whatever the new name is) allows you to do is not really
in the realm of extremely formal clusters-- especially from a company
like Sun which has a serious investment in high-end machines and the
HPC market. If you consider something like the IBM SP/* systems to
be "clusters", then what SGE provides is much much less formal and
organized. On the other hand, the SPs systems could really be viewed
as more MPP systems than clusters because there is a fair amount of
custom support hardware, but I'm not sure we need to go there.
I'd also say that Sun was doing a bit of marketing, and grid has been
a hot term the past few years, so there it is.
It's also important to to realize that if you don't have extremely
strict needs, and your problems are easily adopted to less rigid
environments, then the difference between a cluster and a "well
organized grid" is somewhat minimal. That's a big part of the reason
that stuff like SGE is useful for less complex cluster tasks.
> Maybe "Grid" is just a more common name nowadays than
> "Beowulf" (for marketing reasons) and in the end it's describing the same
> technics (from the point of hardware).
Not really. The concepts of a cluster and a grid *are* different,
although the edges are blurring, mostly because people don't have a
good set of terms for the systems that are in-between-- and we're
seeing more and more of those.
> From the software view, you can have a Beowulf without any queuingsystem
> at all (in the Beowulf book you mentioned I can't find any special
> software for this purpose).
I'm not sure I'd say that. For a Beowulf, there is some assumption
that a single-system view is desirable-- for the user, if not for the
admins and programmers. *Some* queue or DRM system is more or less
required unless you have a very small system that is only used in an
interactive fashion with literally only one user.
"Beowulf Cluster Computing with Linux (2E)" has several chapters on
DRM/Queue systems, including whole chapters on Condor, Maui, and PBS,
along with comments in many of the other chapters. The book
definitely assumes you will have some kind of job management system
on your cluster.
> Just use PVM or MPI (or Linda) and you are the only user on the
> whole cluster.
Right, which is not the way most large compute resources work.
-j
--
Jay A. Kreibich | Integration & Software Eng.
email@hidden | Campus IT & Edu. Svcs.
<http://www.uiuc.edu/~jak> | University of Illinois at U/C
_______________________________________________
xgrid-users mailing list | email@hidden
Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/xgrid-users
Do not post admin requests to the list. They will be ignored.
