On Wed, Jun 07, 2000 at 05:02:11PM +0530, Udhay Shankar N wrote:
> >http://www.jncasr.ac.in/kamadhenu
I can't get to this website. But frankly, this whole cluster business
is getting too commoditized (Anyway, that's the intention right ?). I
guess you saw the $2500 "supercomputer" at purdue running FreeBSD on
slashdot yesterday.
The "8 node Pentium III" part is relatively unimportant. The thing that
differentiates one cluster from another is the kind of interconnect
that binds them together and the bandwidth between the nodes.
In that sense, the "channel bonding" work done by Donald Becker et al
on Beowulf is pretty significant.
Ok, I just got to that page (man, now I better appreciate the meaning of
"slow"). It looks like they're using one card per node, except for the
master which has two. May be that will solve their problem - but it
doesn't make it a "supercomputer".
-Arun
PS: http://acme.ecn.purdue.edu - they did a 5x5 matrix using 10 Mbps cards.
So their cluster will most probably "lose" to kamadhenu in performance, but
probably not in "price/performance"
-----------------------------------------------------------------------
The LIH mailing list archives are available at:
http://lists.linux-india.org/cgi-bin/wilma/linux-india-help
