[Beowulf] torus versus (fat) tree topologies

Joachim Worringen joachim at ccrl-nece.de
Tue Nov 9 19:54:35 PST 2004

Greg Lindahl wrote:
> On Mon, Nov 08, 2004 at 11:45:12PM -0500, Chris Sideroff wrote:
>>  I have had the luxury of testing an SCI 2D torus cluster and found 
>>latency performance to be exceptional.
> Right. So is that a feature of the NIC, or a feature of the torus
> instead of a fat tree/CLOS/CBB network? Answer: the NIC.

Yes, a feature of the NIC, plus: the torus doesn't affect the latency 
significantly. In fact, the cost of a hop in an SCI-based torus, like an 
other reasonable modern torus, is only a low number of nanoseconds (I'd 
have to look it up right now for an exact number which I can not do from 
here; I think there's also such a number in a recent publication on 
BlueGene). A hop in a router node has about the same delay (as it is has 
a similar task).


Joachim Worringen - NEC C&C research lab St.Augustin
fon +49-2241-9252.20 - fax .99 - http://www.ccrl-nece.de

More information about the Beowulf mailing list