very high bandwidth, low latency manner?

[Greg Lindahl]

On Tue, Apr 16, 2002 at 12:24:37PM +0200, Håkon Bugge wrote:

> Also, lack of page coloring will contribute to 
> different execution times, even for a sequential program.

Andrea's kernel patches now have page coloring in them. The code has
lived a tortured life, originally written by the Real World Computing
guys, rewritten by me, rewritten by Jason Popadopoulos at UMd, and
then by Andrea. 3 continents.

> I disagree for two reasons; 
> first, you imply that venture capitalists are naive (and to some extent 
> stupid).

That's what the local Silicon Valley VC tell me about VC. I guess
non-Silicon Valley VC are smarter, then ;-)

> It is to me strange, that no Pallas PMB 
> benchmark results ever has been published for a reasonable sized system 
> based on alternative interconnect technologies. To quote Lord Kelvin: "If 
> you haven't measured it, you don't know what you're talking about".

Maybe that's because other people are measuring their applications,
and not yet another synthetic benchmark? All-to-all isn't interesting
to me. I have plenty of bisection measurements, though, as that's how
I debug Myrinet. Typical variations are around 2%, by the way.

Lord Kelvin engaged in a 10 year flamewar in the Letters of the RAS
against people who thought the Sun was powered by nuclear fusion. He
believed that it was only 10 million years old, and was powered by
gravitational collapse. His mistake was ignoring geological evidence
because he didn't understand it. He probably wrote that quote during
that flamewar. It didn't make him right.

greg