brucec@phoebus.labs.tek.com (Bruce Cohen;;50-662;LP=A;) (02/08/91)
In article <9102041839.AA01139@air.acad> rudy@air.UUCP (Rudy Rucker) writes: ... > I think one of the biggest problems at this point is FINDING > SOMETHING INTERESTING FOR THE ANTS TO DO. The idea of having ants > compete against each other is a suggestive game-oriented application, > analagous to Core Wars. But could the ants ever do something of > commercial interest? Designing the circuits for chips and boards is > a possibility that is sometimes mentioned, but, as I think Hiebeler > pointed out, there are already special purpose programs that do this > much more directly and efficiently. Here's a half-baked suggestion: chemical synthesis. One of the big problems in synthesizing even simple organic compounds is that the yield and overall reaction rate of the synthesis is dependent (usually in a non-linear way) on the yields of the intermediate reactions, and often on the yields of other reactions which inhibit the desired one. Seems like the kind of optimization problem which would yield (sorry!) to genetic search. -- ------------------------------------------------------------------------ Speaker-to-managers, aka Bruce Cohen, Computer Research Lab email: brucec@tekchips.labs.tek.com Tektronix Laboratories, Tektronix, Inc. phone: (503)627-5241 M/S 50-662, P.O. Box 500, Beaverton, OR 97077
burns@latcs1.oz.au (Jonathan Burns) (02/17/91)
Thanks to: brucec@phoebus.labs.tek.com (Bruce Cohen;;50-662;LP=A;) in <BRUCEC.91Feb7133503@phoebus.labs.tek.com> In article <9102041839.AA01139@air.acad> rudy@air.UUCP (Rudy Rucker) writes: ... > I think one of the biggest problems at this point is FINDING > SOMETHING INTERESTING FOR THE ANTS TO DO. The idea of having ants > compete against each other is a suggestive game-oriented application, > analagous to Core Wars. But could the ants ever do something of > commercial interest? Designing the circuits for chips and boards is > a possibility that is sometimes mentioned, but, as I think Hiebeler > pointed out, there are already special purpose programs that do this > much more directly and efficiently. My impression is that ants would be good at tesselation problems. In particular, a memory manager, say for virtual storage, could be built on the basis of static objects (cells) which represent pages in virtual and real address-space, plus mobile objects (ants) which run around linking pages together, promoting virtual pages to real and copying reals back to virtual. The question is: among VM algorithms, are there already cheap ones that get 80% usage or so from real storage? Is it worth investing in CA mechanisms to try and claim another 10%? Answer: maybe so. Inituition says that a bottleneck develops when storage is large, but is managed by a uniprocessor. So, run multiple MMU's (memory management units) in parallel. But now you have to arbitrate communications between the MMUs. So, set them up with a regular neighbourhood topology, and give each a simple set of state rules. Voila, CA. Then, for instance, write a new instruction set for a VLIW machine in emulation. Run a test suite on it repeatedly. Let the ants find their own rules for locally optimum performance. Burn the rules in. A similar, maybe even more suitable application is disk space management for file systems and databases. A pleasant aspect is that cached storage is a modular part of any system. Build a cache with CA smarts, and it doesn't entail any change in CPU architecture. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Jonathan Burns | Clashing for the warrior, whose strength is not to fight burns@latcs1.lat.oz.au| Clashing for the refugee, on the unarmed road of flight Computer Science Dept | And for each and every underdog soldier in the night La Trobe University | We gazed upon the chimes of freedom flashing -Dylan ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~