[Server-sky] List looks good!!

[M. Edward (Ed) Borasky]

OK ... looks like we are live. So ... to continue the discussion Keith
and I started at NedSpace:

My first question was, "What instruction set architecture?" Now, if
Server Sky were to be deployed in the next six months, I think that
would be obvious -- x86_64 if High-Performance Computing / floating
point number crunching was the target market, and maybe ARM if there
was no need for floating point.

But now, let's suppose this is a couple of years out. Do we really
expect that x86_64 architecture to have mutated again to, say, a
128-bit form? And do we think the "more advanced software" we expect
to have then will run efficiently on a hacked-up x86_64? For example,
Intel and AMD had to put hacks in the architecture to support
virtualization, after having to put hacks in for 64-bits, after having
to put hacks in for PAE, multimedia, etc.

It's hack upon hack at this point in time. *Some* of the hacks were
clean, but some of the hoops that Xen has to jump through on x86 even
*after* the virtualization hacks went in are just not reasonable. I
think, as long as we're dreaming, we might as well dream about an
architecture that *doesn't* make the operating system and compiler
designers and performance engineers have to strain their brains.

I'm starting to think the IBM / Sony / Toshiba "Cell" processor just
might be better for the kinds of jobs a Server Sky would be
performing? I suspect the real answer is, "depends on who's paying the
bills." :) But I think if I were to pick the architecture more likely
to be "cleanly 128-bit and virtualizable", it would be the Cell.

Comments??

-- 
M. Edward (Ed) Borasky
http://www.linkedin.com/in/edborasky

I've never met a happy clam. In fact, most of them were pretty steamed.