Dan Clemmensen wrote:
>
> my inner geek wrote:
> >
> > Dan@Clemmensen.ShireNet.com (Dan Clemmensen):
> >
> > URL re: Linux clusters
> >
> > http://www.globec.com.au/LDP/HOWTO/Parallel-Processing-HOWTO-3.html
> >
> > -------------------
> > geek@ifeden.com
>
> Interesting, but fairly basic. I should tell you a little about myself.
> I've been a data communications programmer for over 25 years.
> In 1985, I was one of the founders of the company I still work for.
> We've been building peer-processor datacomms swithces since I designed
> the architecture for our first one in 1985. I do try to keep up with
> this stuff. Please beleive me when I say that the networking isn't
> particularly hard. The problem is in task assignment and task migration.
>
> Note that when the computers are all in the same room, there are cheaper
> ways to build a Beowulf than by using ether-switches. In particular,
> if the comuters are arranged in a grid, each computer can use two 100BaseT
> cards instead of one. In this arrangement each computer can act as a crosspoint
> for its row and column, switching on average only as many packets as it
> originates or terminates. This gives you massive, incrementally expandble
> bandwidth for approximately $50/node.
I too have been an Applications, Communications, OS, 4GL and AI programmer/licensor for the last three decades and invite you to look at my sole remaining uploading project.
Please, have a look at Ingrid because there is this really novel idea about how to have Ingrid running in up to 12 windows on each computer. I like the idea of each computer acting as a crosspoint for its row and column as this fits with my concept of how to assign a task to a dormant Ingrid processor in a mesh of Ingrid resources. As a single Ingrid processor is in effect a primitive AI inference engine capable of neuronally re-plasticizing migrating features, it has long been my dream to have a parallel version of Ingrid running on something like a Beowolf cluster. Right now I'm putting together a team to consider the variety of 'skins' that will be needed in the various user interfaces. Whether it be talking pens or a replacement to the money system, Ingrid is a 'must be considered' front runner in terms of the small amount of effort to get it going. Unfortunately the supercomputing community is very shy about their resources being taken over by Ingrid.
The pre re-write Ingrid source code is open for you to see how well the nodes can be made to run in under 50Kb of integer programming (as a subconscious processor without a UI.) Ingrid survived for me because of its small size whereas my 50Mb Global operating system/4GL died in a McMafia court battle.