HOWTO discriminate switches?

[Greg Lindahl]

On Mon, Sep 17, 2001 at 02:12:02PM -0500, Richard Walsh wrote:

> Without trying to be a salesman for SCI, they claim their 66/64 PCI
> cards to be 1.33 G_BYTES (not bits) per second bi-directionally with a
> 1.46 microsecond latency.

Right. This is not a measured value for user process to use process
communication through the PCI bus, which is what users care about.

> I do not know about the _REAL_ math, buy an anlysis that I did for
> a 512 node system a year ago comparing switched versus non-switched
> topologies based on SCI cards showed the SCI (cards only) configured 
> system to be cheaper than Myrnet

You should work out the bisection bandwidth. It is substantially lower
for SCI for large systems, while Myrinet's bisection scales linearly.
This is why all large clusters which need low latency and high
bandwidth are Myrinet and not SCI.

> Baseline performance of the network had to be 
> close to that of the Cray T3E.

And if you work out the numbers, neither SCI nor Myrinet can match a
T3E, especially on latency.

greg