Author: Robert Hyatt
Date: 20:49:25 09/16/01
Go up one level in this thread
On September 16, 2001 at 21:59:46, Vincent Diepeveen wrote: >On September 16, 2001 at 09:41:29, Robert Hyatt wrote: > >>On September 16, 2001 at 06:02:24, Tony Werten wrote: >> >>>On September 15, 2001 at 22:34:27, Robert Hyatt wrote: >>> >>>>On September 15, 2001 at 03:28:18, Tony Werten wrote: >>>> >>>>>On September 14, 2001 at 22:56:06, Pham Minh Tri wrote: >>>>> >>>>>>I see that dual computers are expensive, not easy to own and still limited in >>>>>>power of computing. >>>>>> >>>>>>I wonder how good / possible if we use all computers in a LAN for chess >>>>>>computing. LANs are very popular and the numbers of computers could be hundreds. >>>>>>Even though a LAN is not effective as a dual circuit, but the bigger number of >>>>>>processors could help and break the limit. >>>>>> >>>>>>What do you think? >>>>> >>>>>When you search a chesstree, a lot of times you come into parts of tree that you >>>>>have searched before. You either don't want to search this part again ( you have >>>>>searched it deep enough before ) or you want to have the best move from the >>>>>previous search. Hashtables do exactly this. >>>>> >>>>>In a LAN (or a cluster) you don't share this hashtable and therefor are >>>>>searching the same tree (or parts of it ) time and time again. If you count the >>>>>number of nodes searched per second it's a linear speedup but effectively it's >>>>>useless. You have to add a lot of computers before you get any real speedup, >>>>>specially in the endgame. >>>>> >>>>>cheers, >>>>> >>>>>Tony >>>> >>>> >>>>This is not necessarily true. Several programs have distributed the hash table >>>>across network nodes. It requires small changes to the basic search algorithm, >>>>but a distributed hash table is not only doable, it has been done more than >>>>once. >>>> >>>>I will probably do this in the distributed Crafty when I do it... >>> >>>I guess sharing the first x ply on a 1 or 10 Gb network will work, but I don't >>>think you can use the normal dynamic tree splitting. I gues you have to split at >>>a static depth ( decided in the first search ) ? >> >> >> >>That is not my intent. Of course, I won't try to split at the deep levels in >>the tree that I split at now. But for (say) the first 1/3 of the plies in the >>current search, splitting is certainly doable. This will just be a tunable >>parameter that will have to be adjusted depending on the hardware and network >>speeds. >> >> >> >> >>> >>>Just to get an impression. How many single Xeons do you think you'll need to get >>>the same speedup you get on a quad Xeon ? After that, does it scale ? >> >> >>I can't imagine that this will be less than 50% effective. Or, if we take deep >>blue as an example, no less than 30% effective. I would think that 8 cpus would >>be very close to the quad... >> > >Bob, you're pretty optimistic here i think. > >A quad xeon delivers 3.1 > >Now how many single xeons you need for 3.1, well you lose like 50% just >like that everywhere we still didn't talk about speedup. then you lose >another factor of 6 or so because you can't hash last few plies. You make too many assumptions. The 50% "loss" is not a given. I don't accept such a number with no basis or experimental proof. And there is a lot of experimental evidence that it is wrong. I don't accept the "factor of 6" about hashing the last few plies either. I haven't seen the tests, but I do recall your saying "not hashing the q-search is real slow" yet for me it works fine. I don't see a thing that says that not hashing the last few plies is bad. At with a latency that can be down in the .5 microsecond range, that kind of network speed is comparable to the nps value for a fast engine. Which means it _could_ hash in the q-search and keep up if needed. > >Now this factor of 6 is the biggest problem. the system time already isn't >the problem. So i would say 3.1 x 6 = 20 nodes at least. Why not make it a factor of 100? Or 1000? I don't particularly like to base performance guesses on wild random numbers... > >O yeah and of course at least a 1.25 gigabit/s network. At a 100mbit >network i doubt one ever gets a positive speedup anyway > I offered to make a wager. +I+ will get a speedup on a 100mbit network either this year or next. Guaranteed... >> >> >> >> >>> >>>cheers, >>> >>>Tony
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.