Author: Robert Hyatt
Date: 20:59:26 10/03/01
Go up one level in this thread
On October 03, 2001 at 17:04:25, Miguel A. Ballicora wrote: >On October 03, 2001 at 11:44:30, Robert Hyatt wrote: > >>On October 03, 2001 at 10:51:13, Miguel A. Ballicora wrote: >> >>> >>>There is between two separate searches through the hashtable, killers etc. >>>That is how "synergy" lowers the time in iterative deepening. >>> >> >> >>Ok.. You do a 5 ply iteration on one processor, and a 6 ply iteration on the >>other, in parallel, and show me the "synergy". Notice that your example is >>_exactly_ my point. "SERIAL". The 5 ply search _must_ be done first. It >>then orders the 6 ply search more efficiently. If you do them in parallel, >>you don't get this synergy. >> >>And actually, "synergy" is the wrong term here, because iterative deepening >>is _also_ a purely sequential algorithm. > >So? synergy applies to things that are sequential too. > But that has nothing to do with "parallel synergy". The argument is for or against super-linear speedup. To even come close to justifying such a claim, there must be some sort of parallel synergy exhibited. And there isn't, in alpha/beta. But even if there were, the two-thread on one processor algorithm would be faster than the normal one processor algorithm, and the super-linear speedup disappears that way too... >>>Yes, really. On other planets too. I am not going to discuss more on this >>>example because we are wasting time, yours and mine. It supposed to illustrate >>>an aspect of synergistic effects and nothing more. There many examples of this >>>in science. I understand you might not believe what I tell you, so please >>>believe any textbook that deals with friction and mechanics. Those are typical >>>examples. >> >>I have one right here. Which part would you like to discuss here? It is >>_always_ easier to drag than to lift and carry. Perhaps in "dragging" you have > >Well, it is not _always_ easier and explained why already. If you want to find >it in a book it may be in the exercises of the chapter that deals with static >and dynamic friction coefficient. Anyway, this example was colateral. > >>to provide a lifting force to overcome part of the friction. >> >>> >>>Faster solution does not involve expending more energy. Do not confuse >>>kinetics and thermodynamics. In fact, life is finding faster ways >>>to do things that might not look the most favorable (energetically). >>> >> >>That is _my_ point. But in your example, it _does_ involve expending >>more energy. One man can drag 10 200lb blocks N feet in M units of >>time. Two men can do that in M/2 units of time. If both drag a block >>together, they will expend more energy than if each drags one. Because >>the acceleration is not free and they have to move _twice_ as fast. And >>that means more "work" is being done. A computer can't do "more work". >>So the example simply doesn't apply. >> >> >>>You did not like Bruce's example? This one does not involve scientific >>>knowledge: >>>Team A: One person , 1 cart, 100 boxes too heavy to lift by only one person. >>>Team B: Two persons, 1 cart, same 100 boxes. >>>Task = move them 100 yards. >>> >>>Team B will take <50% of the time that it will take for Team A. Lifting >>>once onto a cart is easier than pushing. >> >>I use an inclined plane to slide the boxes onto the cart. Remember, if you > >You do not have an inclined plane, that was not part of the example. >If you change the rules of this simple example we can be discussing >for ever going for branches of the "analogy tree". I knew this is what >could happen. We end up discussing things that are irrelevant. > Then the speedup is infinite, and the example is no good. Because _one_ man simply can not do the job at all. This has nothing to do with alpha/beta, because we _know_ that one processor _can_ search the tree. >Miguel
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.