[Beowulf] Cheap SDR IB

[Håkon Bugge]

Mark,

At 15:09 31.01.2008, Mark Hahn wrote:
>I did not claim the opposite - I said that for small, cost-sensitive
>clusters, it would be unusual to need IB's advantages (high bandwidth
>and latency comparable to other non-Gb interconnects.)
>
>in particular, I'm curious about the conventional wisdom about weather codes
>and bandwidth.

k

>I was curious about this: you only used one DDR port; was that because
>of lack of switch ports, or because WRF uses bandwidth <= DDR?

The system is a general purpose benchmarking system; not particularly 
crafted for running WRF. Based on a slightly apples-to-oranges 
comparison, you will see that QLogic's SPEC MPI2007 submission 
contains a WRF number (374s) which is _very_ similar to what I 
reported. This is an indiction that WRF on this system / dataset is 
not restricted by SDR bandwidth (also, for the record, this is a 
slightly mix of compilers, Pathscale 3.0 and Intel 9.1, - but 
they  both do a decent job on WRF).

>sure, and these are very fat nodes for which a fat interconnect is
>appropriate for almost any workload that's not embarassing.  but really
>I wasn't suggesting that plain old Gb (bandwidth in particular) was
>adequate for all possible clusters.  I was questioning whether IB 
>was a panacea for small, cost-sensitive ones...

I do not agree that  dual-socket, dual-core Woodcrest nodes these 
days are "very fat". A quad-socket, quad-core is. A quad-socket, 
dual-core or a dual-socket, quad-core might be considered semi-fat...

Hakon