Gigabit Ethernet switches and network adaptors.

['math at velocet.ca']

> The NS based stuff is low-end and cheap - you get what you pay for - but the
> drivers are rock solid and provide a good way to get started with GigE.
> 
> On a high enough powered box, you might even get decent throughput but it's
> at a cost of cycles.


Here's an interesting question: even with the fastest network interconnects
(SCALI, etc), we dont see 100% scaling at large numbers of nodes. There
is some free CPU left over. So what if you had slightly less efficient
equipment? I realise it would cause a slowdown for sending out messages
as the latency may be increased, but if the latency is the same as 
for the high end network equipment and only costs more cycles, is it
conceivable that the scaling and performance of this cluster with slightly
less efficient equipment would be similar? (Again, there's a big assumption
here that we can find such equipment that has the same latency when
extra cycles are involved, which may be the source of much latency for
many cards in the first place).

/kc