Re: Hardware Cost

From: Brian Atkins (
Date: Wed Apr 25 2001 - 13:13:53 MDT

James Rogers wrote:
> At 09:53 AM 4/25/2001 -0700, Dani Eder wrote:
> >[...snip...]
> >I'm curious how this type of Linux cluster compares
> >to the processor requirements for a Web-Brain
> >implementation or an estimated size for a seed AI
> >(if anyone has taken a stab at that yet).
> >[...snip...]
> >The cost breakdown is as follows. Note the cluster
> >has 64 compute nodes + 2 supervisory nodes. Costs
> >are current ones from
> As a general comment, you don't seem to be considering that these types of
> clusters are tuned to solve very specific problems. Their solution isn't
> necessarily your solution. The choice of hardware is not arbitrary;
> interconnects, RAM, switching, and proc/mobo hardware is all carefully
> chosen to maximize throughput for the problem in question. A seemingly
> trivial mis-design in your cluster can cost you an order of magnitude in
> performance.
> Some problems can be solved very cheaply in clusters, with almost perfect
> linear scaling (and occasionally, the rare super-linear scaling). Other
> types of problems demonstrate logarithmic scaling, even with very expensive
> interconnects. I expect the AI problem falls somewhere in the middle,
> being neither embarrassingly parallel nor excruciatingly serial.

James I was looking at the Blue Gene design in more detail (do you know of
anywhere that has more info than the IBM site?) and it actually looks like
it might run in a very similar fashion to a Beowulf. Each chip might be
thought of as 32 nodes in a Beowulf connected together to a switch, or
possibly each chip could be thought of as a single node capable of running
(32 x 8) threads at once using 16MB ram. Since the CPUs will incorporate
simultaneous multithreading they should be able to keep pretty busy even
when some threads are waiting for memory access. The nice thing is that
this design should nicely allow for a lot more embedded RAM per CPU in the
future for Blue Gene 2 or whatever as the transistor size decreases in the
second half of the decade.

What I don't understand is: will there be some kind of micro-OS running
on each of the 32 CPUs per chip, and how much of the 512k RAM/CPU would
it take up? I get the impression that is how it will work since their site
talks of using message passing for inter-CPU communication. Supercomputer
operating systems are something I've never learned much about, and I'm
curious if they are significantly different from normal.

<good Beowulf advice snipped>

Brian Atkins
Director, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT