> It was thus said that the Great Sam Ismail once stated:
> > On Mon, 25 Jan 1999, Max Eskin wrote:
> > > On Mon, 25 Jan 1999, Sam Ismail wrote:
> > > > Yes but we're using those for a VCF project.
> > > Using 80 C-64s in a single project? Heck, I'm curious...
> > Were basically in the preliminary feasibility study stages of building a
> > massively parallel vintage computer. The point? To demonstrate that old
> > hardware that can be picked up for pennies can be combined to attain
> > amazing amounts of computing power.
> The biggest problem will be communication speed between nodes.
Depending deeply on the structure of the task.
> > Hans Franke is the project leader. Currently we are talking about the
> > architecture of such a thing. The current wisdom from Hans is to use 1541
> > drives instead since they would be easier to network and are self-booting,
> > not to mention plentiful and cheap.
> And you'll end up with a ring architecture, since if I remember correctly,
> each drive only had two IO ports on it. That is, if you don't hack some
> hardware to get a parallel connection to each computer (like 10BaseT hubs
> today). And you'll need to select your problem domain to complement the
> speed issues you'll have in communication, or you may end up with something
> that's slower than a single C-64.
Redarding th information Salam gave, your idea might be right, but
as you said, its also depending on the problem to solve. Also you
just imply that the conectionstructure will be a simple brute force
bus ... Just wait (or join the team).
> -spc (Let's see ... 80 machines in a ring configuration, average
> distance between two nodes is 40 hops, and at 9600bps
> (assuming the hardware can do that) you have one bit taking
> 104 uSec between nodes (with an additional .001 uSec per
> foot, propagation delay you know 8-), so you have 1.04 mSec
> per character per hop. So to send a one byte message to
> an arbitrary node will take, on average, 40 mSec. Round trip
> will therefore take nearly 1/10 second (and that's assuming
> no delay between switching nodes---YMMV in actual use ... ))
a) what if there is no peer type network, but rather a structured
hirarchy of domains ?
b) what if there is a bus rather than a ring (with switching time)
c) What if the speed in the serial part of the network is higher ?
d) what if the amount of date transfered includes low overhead ?
Come in and join the definition team.
Gruss
H.
--
Ich denke, also bin ich, also gut
HRK
Received on Wed Jan 27 1999 - 07:09:25 GMT