Slashdot Mirror
Looking Back at 1984 Report On "Radical Computing"
An anonymous reader writes "The Department of Defense has just released a long restricted report (PDF) by the JASON group entitled Radical Computing. This 1984 study outlines a number of alternate computing methods that could 'result in a radical improvement in computing.' The study attempts to explain the paradox of how the Russian lag in developing VLSI chips curiously did not critically hinder their accomplishments in space missions, ICBMs and chess computation. The authors speculate that the Russians might have achieved breakthroughs in alternative computing methods such as residue arithmetic and symbolic computing. (More cynical types assume the Russians bought or stole US chips from the French or other too-helpful go-betweens.)"
"The paper, published by the Government Attic website, also mentions how, eventually, highly parallel computers could make use of these alternative computational methods. Also discussed are such things as functional programming, interval arithmetic, recursive machines, multiple processor concurrency, fast recurrence evaluation, DDA machines, data-flow, and hyper-column cortex model. Which of these ideas ever came to fruition?"
In soviet russia.. Radicals compute you!
I' trying to remember who said this. But during the Cold War, the intelligence folks got so paranoid that they were attributing things and capabilities to the Russians that, after the Cold War ended, the Soviets were no where near having any sort of capability or had any sort of plans. One of the more well known over estimation was Soviet military capabilities. When the Cold War ended, the intelligence community couldn't believe how far off they were - most of there "insights" were over active imaginations.
RIP America
July 4, 1776 - September 11, 2001
One of the main serious uses of computing, especially in the cold war, was solving partial differential equations. Whether these be for orbital calculations, stability analysis, EM simulation, etc..., solving partial differential equations is a critical part of any advanced engineering program.
The American approach really started in the 50s with the advent of programmable computers, and is very stereotypical: just find a decent approximation. Modern western engineering is all about using pretty advanced computers to find arbitrary numerical approximations to tricky PDEs. It's reached its culmination in modern engineering design, where most advanced products are designed and simulated in computers, and prototyping only occurs at the very end of the process.
The Soviets had computers.... some home built, some Western, but generally speaking they weren't very good. The Soviet approach was also very stereotypical: get an army of mathematicians and engineers to find exact analytic solutions to the problems you're trying to solve. You'd have armies of engineers and technicians designing things that in the west we'd give to a couple of engineers with some computer time.
The end result is that some Soviet engineering is stunningly brilliant. And a lot is absolute crap. One of the reasons the west won the cold war is that we were just much better at solving partial differential equations. This report is unsurprising... the Soviet approach just seems so stupid to any Western engineer unfamiliar with it, that you'd have to assume they had some magic trick up their sleeve. But nope, just a lot of brainpower misdirected into a lot of horribly inefficient pursuits.
Thank you for your comment, Sznupi, but I have a few remarks, that I may make because I am myself a computer designer. The Setun example was served to me several times in the last decade and the same arguments apply :
1) Just look around you : where does ternary logic live ? in some Russians' fond memory. OK.
Show me where ternary logic can replace things : AFAIK, it is used in *some* multiply hardware, under the name of Booth recoding.
http://en.wikipedia.org/wiki/Booth%27s_multiplication_algorithm That's all, and it is not always practical : booth computations are a bit faster but recoding is a pain.
http://www.fpga-guru.com/multipli.htm
2) do the maths : Ternary logic values on binary wires :
- either you use 2 wires to encode 3 values and you lose 1/4 of the coding space (as in any base conversion)
- either you use the 3-wire 1-hot encoding and... well, you win nothing.
Now imagine you have binary memory : you lose 1/4 of the capacity. You can recode data so you lose less, but the less you lose in space, the slower it runs because it adds complex base-conversion circuits, with all the carry chains and the likes.
Memory in the first Russian ternary computer was certainly magnetic core memory : with the epoch's electronics, it was not difficult to encore magnetic 3 fluxes. But it does not work well in today's very high speed logic, where noise resilience and process variations can kill electric margins.
Conclusion : we live in a binary world, it's not by mistake.
Now if some electronic circuit worked WELL in ternary, it would not be enough : it would have to work WAY BETTER than today's binary circuits to even consider acceptation.
Don't get me wrong : I respect Russians a lot. But we all make mistakes and invent our little prides... All engineers have their failures... It's part of our learning. It is a greater failure to not learn from our mistakes.
Ternary computations were a "local minimum" for a given time and technology. And I don't regret the time when the US's supercomputers (http://en.wikipedia.org/wiki/CDC_6600) used 60-bit words and 6-bit bytes. The next generations of Cray designs went to 64-bits wide registers and 8-bit bytes, and they even adopted (reluctantly) IEEE Floating point numbers. This proves that even when technical merit is stellar, it is useless (and even laughable) if it can't interface to the other computers. Adapt or die bragging.