In a parallel machine with many thousands of processors the routing of information between processors is a key task, which turns out to require as much hardware and perhaps more sophistication than local computing itself. There are at least two basic engineering solutions to the routing problem: one followed by most research projects is of the "packet switching" type, that behaves as a mail service, with data carrying addresses to route the packet through the system. The other, more similar to a traditional telephone system, has connections made and broken (or enabled and disabled) as required for exchanging information. These solutions, based on silicon technology and digital electronic, may be quite different from the routing solutions used by the prototypical parallel machine — the brain. This paper asks questions concerning routing information in parallel machines with an eye to biological wetware. It is divided in four disconnected parts, that do not contain finished results but consist of suggestions for future speculations: 1) How to make Infinity Small. 2) Routers and Brains 3) Classifying Parallel Machines 3) The Problem of Remapping
This working paper is has been brought to you by the modern wonders of microcassette dictating equipment, through which Professor Poggio can now cough up working papers while doing something else more important.
MIT Artificial Intelligence Laboratory
MIT Artificial Intelligence Laboratory Working Papers, WP-258