🚀 go-pugleaf

[Original article deleted due to too long lines]

How I want it done

Modify an ordinary alpha-beta searcher like this:

*  Make sure every modification use at most epsilon/3 cpu

*  Make the program use a hash table (if it already does
   then this step is not neccesary)

*  Make the program maintain a semi-ordered lossy list
   of the positions that have the greatest expected
   evaluation time. (send)
   [position, start-time, end-time, alpha, beta, depth]

*  Make the program merge information sent to it with
   the hash table. (receive)

This will make the program at most 1/(1-epsilon) times slower.

Then write programs that will read the list of positions and search
them in parallel, if they are few, making each of them work like
the first (computer #n+1 computer aiding computer #n) might give
acceptable results.

That is, if you don't aim for linear speedup.

To make a program modify it's current search take some more
changes but need not slow it down. (hard to write, easy to run)

-- 
Urban Koistinen - md85-epi@nada.kth.se