[Beowulf] Network considerations for new generation cheap beowulfcluster

Eugen Leitl eugen at leitl.org
Tue May 22 06:22:56 PDT 2007


On Tue, May 22, 2007 at 09:02:38AM -0400, Larry Stewart wrote:

>    I haven't heard of anyone using a diamond lattice to wire a cluster.
>    We're
>    using a Kautz graph, which already makes my brain hurt.
>    What would the advantages of a diamond lattice be?  In terms of

4 is the smallest number of edges for a node to span a 3-d mesh,
without using a switch. It is suitable for problems with a purely
local communication pattern only.

6 edges/node would be better, since resulting in a cubic primitive
lattice of nodes (3d torus, or similiar), but I haven't seen
a low end box with more than four GBit NICs. Speaking of which,
why is nobody integrating small (8-port) GBit switches on motherboards
themselves? I've only seen this in firewalls so far.

The disadvantage of a diamond lattice is that decomposing your
physical system is more complicated (I don't even know which 
polyhedra and common faces one would wind up with). It is probably
not worth the hassle.

>    bisection and diameter?
>    Ease of wiring?


-- 
Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
______________________________________________________________
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE



More information about the Beowulf mailing list