OK. I buy your comparisons. And it had gone past me that Xgrid is a
grid, and not a cluster-in-the-Beowulf-sense. That's a really good
distinction.
There is one limitation though - Rendezvous. If you look at that, then
an Xgrid has more a traditional cluster network topology. All the nodes
are connected on a single subnet. This is different from something like
Globus, which is designed to leverage the internet or WAN's as a
communication medium. I guess that that says that its not about the
network topology, and the network itself being trusted, its about
guaranteeing that a result comes from who it says it does - "formal
membership" as you said below.
I've never looked at it, but I would be interested to see if Globus has
a "Signed Results" API to protect against forgery.
On Jul 27, 2004, at 11:30 PM, Jay A. Kreibich wrote:
Most of this definition comes from the book "Beowulf Cluster
Computing
with Linux" and its sister book "Beowulf Cluster Computing with
Windows", which are out-growths of the book "How to Build a
Beowulf."
If you read through the long list of authors for these books, you'll
see a lot of names that are pretty core to low-cost cluster
computing,
so I would tend to follow their lead.
Linux seems to be extremely common, but Windows, Mac OS X, or any
other "consumer desktop" OS is acceptable. Some would claim Linux
is
the only true system, although most of the people saying that are
Linux types who think anything UNIX-like was invented by the Linux
community. Similar comments could be made about the system having
x86 hardware of some flavor. If anyone else has a definition,
please
throw it out (along with a source, if you have it)-- I love to hear
stories.
Well, as a Linux fan, I hope that I understand the history of what I
use day to day. Strength is where you find it, it was fairly trivial
for people to modify the kernel in Linux to do things like a shared
process space, so that is why you find people insisting that Linux is
the way to go. But this will change, as strengths of other systems make
themselves apparent. Apple has always placed a high value on ease of
use.
I have an experimental cluster in my base with 4 nodes made out of old
486 boxes. If I count the number of hours I spend just making sure that
things configured right I might realize that I need a life. Apple has
taken all of that and reduced it to zero.
Anyways, If we stick with that definition for now, then most Mac
clusters
are "Beowulfs," the exception being if you use a
high-speed/low-latency
interconnect like Myrinet or InfiniBand.
Or, dare I say it?! Gig-E...
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.
Formal membership, centralized administration, fixed topology, and a
lot of other things account for the difference between clusters and
grids. Others can provide more discussion if you wish.
Is there MPI for xgrid?
Yes, but I'm fairly sure it is only on the OS X side of things. See
the release notes.
-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.
[demime 0.98b removed an attachment of type
application/pkcs7-signature which had a name of smime.p7s]
_______________________________________________
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.
