[Beowulf] Network considerations for new generation cheap beowulfcluster

Robert G. Brown rgb at phy.duke.edu
Tue May 22 06:37:32 PDT 2007


On Tue, 22 May 2007, Larry Stewart wrote:

> What would the advantages of a diamond lattice be?  In terms of bisection and 
> diameter?
> Ease of wiring?

Four ports per system, probably, in a 3d lattice.  3d is good because
the volume (number of hosts) scales like the maximum number of hops
between hosts cubed.  With two ports you can make a ring but max_hops
increases linearly with number of hosts.  With three ports you can make
a triangular lattice or a tree structure depending on whether you want
to optimize topdown or peer-to-peer, IIRC.  With four you can make a 2d
square lattice where the volume scales like the square of max_hops OR
you can make a tetrahedron and get a volume that scales like the cube of
max_hops.  Probably with toroidal closure, although, as you say, it
makes one's head hurt to visualize it and building the routing table
would be "interesting".  In fact, you'd probably have to write software
to build the per-node routing table.

I built a small demo cluster with a tetrahedral lattice of interconnects
like this ONCE for Linux Expo about seven or eight years ago.  Once (as
Voltaire famously replied to the Marquis de Sade) is philosophy...

    rgb

>
> -Larry
>
>

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu





More information about the Beowulf mailing list