Date: Thu, 27 Jun 2002 14:27:42 -0700 From: Terry Lambert <tlambert2@mindspring.com> To: Jonathan Lemon <jlemon@flugsvamp.com> Cc: Julian Elischer <julian@elischer.org>, Greg 'groggy' Lehey <grog@FreeBSD.ORG>, arch@FreeBSD.ORG Subject: Re: Larry McVoy's slides on cache coherent clusters Message-ID: <3D1B834E.70573706@mindspring.com> References: <Pine.BSF.4.21.0206271044050.69706-100000@InterJet.elischer.org> <3D1B7391.38F10284@mindspring.com> <20020627152602.A1020@prism.flugsvamp.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Jonathan Lemon wrote:
> On Thu, Jun 27, 2002 at 01:20:33PM -0700, Terry Lambert wrote:
> > My personal target rests above NUMA, where there are relatively
> > glacially slow communications channels, compared to CPU speed;
> > this is basically the environment in which, for example, you
> > have literally millions of processors operating from incomplete
> > information with potentially lossy communications channels.
>
> A.K.A. "The Internet".
Actually, a worse communication vs. CPU speed ratio than that,
eventually. There are a couple of problems that can only be
solved with an architecture that assumes a progression of the
ratio.
I think Larry persuasively demonstrates that there is a hierarchy
in communications channels vs. CPU speed that is not accounted for
in most OS design. My scale ("Lambert's Interconnection Scale"? 8-))
would be:
---- ---- ---------- -----------------------------------
CPUS DIES SEPERATION NAME
---- ---- ---------- -----------------------------------
1 1 0 Processing (8-))
N 1 0 SMT
N M 1 SMP
N M 2 NUMA
N M 3 Distributed (full information)
N M 4 Distributed (partial information)
N M 5 Distributed (partial functionality)
---- ---- ---------- -----------------------------------
The hardware DES breaker that was built as a proof of concept was
purpose-built hardware with a seperation of 2.
The 65,536 processor machine that Good Year built for modelling
laminar airflow on the full shuttle airframe was purpose built
hardware with a seperation of 2. So were most of the Connection
Machine series from Thinking Machines, Inc..
SETI@Home is a purpose-built machine with a seperation of 3, and so
are the protein folding and crypto-breaking and similar systems.
The Javalin research project was a virtual machine general purpose
computing platform with a seperation of 3.
One of the things that Kazaa is attempting is to build a general
purpose computing platform -- a real machine -- with a seperation of
3 (whether they realize this or not is another matter).
Larry's presentation claims (in slide 11) that the traditional MP
approach has been to "build a solution, and scale it up", e.g.:
``We're at 2, can you get to 4?''
``We're at 4, can you get to 8?''
Etc.
And then he asks:
``Can you go 3 orders of magnitude farther?''
Which is maybe the wrong question; it pegs your initial position
farther out on the incrementalism scale, but doesn't answer the
question of how to get from N to N+1 for an arbitrary N.
-- Terry
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-arch" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3D1B834E.70573706>