On Wed, Jul 28, 2004 at 07:41:06AM -0700, David Leimbach scratched on the wall:
> I would probably put it to you that all grids are clusters but not all
> clusters are grids.
I would actually say the reverse is true:
- A cluster can pretend it is a grid
- A grid cannot easily pretend it is a cluster
Grids are generally considered to be less organized than clusters
(in the sense of different hardware, different admins, different
environments, different organizations, different networks,
different topologies, etc., etc., within the same grid, while a
cluster would have all of these the same).
If you have software written for a grid that makes few assumptions
about these kinds of things, that software will likely still run on
a cluster-- and run quite well. On the other hand, if you have have
software that was written for a formal MPI cluster (like VT's "X") that
has strict requirements on compute environments, network
interconnects, bandwidth, and latency, it will likely flop over and
die if you try to run it on a typical grid-- if you can get it to run
at all. It might very well run faster on a single machine.
In some ways, "grids" are really a hard look at what a traditional
cluster is, and what can be thrown away from a formal cluster while
still allowing useful work to be done for many parallel applications.
In that sense, clusters are kind of a super-set of grids (in
capabilities and cost).
Besides, if all clusters were grids, it wouldn't be the "hot new
thing" it has been for the past few years. Clusters have been around
for decades and decades.
-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.
