Author: Robert Hyatt
Date: 10:17:40 09/03/03
Go up one level in this thread
On September 03, 2003 at 12:45:40, Vincent Diepeveen wrote: >On September 03, 2003 at 12:34:53, Sune Fischer wrote: > >>On September 03, 2003 at 12:23:08, Vincent Diepeveen wrote: >>> >>>The only one who you are confusing is yourself. >> >>Correct, I'm confused why _you_ think threads doing independent searches needs >>to communicate all the time. >> >> >>>DIEP runs fine at any latency, but the speedup simply gets a lot less when the >>>latency goes up. >>> >>>There are many practical problems. >>> >>>You speak about shipping messages. >>> >>>When are you going to receive them. Check each millisecond? >>> >>>Or let the OS decide? >> >>I think I explained this already in other parts of the thread. >>You can check as often as you want, it's sending the message that's expensive, >>not checking for it. >> >>>The OS fires at 100Hz, so things like processes that are sleeping because of the >>>OS putting them to sleep (when locking and for 600 times they can't get the >>>lock) then you have a latency of 10 ms before the process is awake. >>> >>>You are aware of such problems? >> >>Yes, but this is concerning spawning and killing searches, not communication >>between active search threads, don't confuse the issues. >> >>>DIEP only can run well parallel at a shitload of cpu's thanks to statistical >>>chances that a scenario X doesn't happen much. >>> >>>That took 1.2 years fulltime work. Still tuning some details. >>> >>>There is a lot of communication. It is very easy to test this yourself. >> >>So give me an example, don't blow smoke. > >ok easy example. > >Only YBW works well. The other parallel search ideas you can shredder. > >Everyone has physically found this out. It is the only algorithm which doesn't >blow your speedup. > >Feldmann found out. > >Hyatt found out. > >Diepeveen found out. > >And i bet many others. Of course i tried other ways of splitting. They suck for >speedup reasons. They suck terrible, then i say it very politely still. So i >also wasted time at finding out that YBW is something you MUST use. > >I assume you know what YBW is. If not there is plenty of articles describing it. >Some are downloadable for free if you are not a member of ICGA journal. I am >sure Hyatt has described it as well. > >So when at 500 cpu's, diep first starts search at 500 Mhz cpu and the other 499 >idle. > >Now i ask you, do you want to let it perform the entire search at 500Mhz, or do >you prefer knocking in another 499 cpu's? > >If so how to do that without a lot of communication? > >>>Just get a cheap network card. Say 100mbit and connect 2 pc's. Now let them do a >>>parallel search. >>> >>>Please report back to me when you have a speedup > 1.0, because initially you'll >>>be slower than 1.0 i bet. >> >>I bet not ;) > >Your first year of versions i am pretty damned sure it will be < 1.0 :) > >I remember paderborn 1999. DIEP when run at Hyatts quad 400Mhz box was 6 times >slower than single cpu 450Mhz PII. you can split the tree _only_ at the root, and with two processors get a 1.5x speedup. This has been done. I did it in 1983. Of course, with 4 processors, the speedup is still around 1.5x. But you said two. > >Of course in those days GCC sucked more ass than it does now, but still factor 6 >:) > >Directly after tournament i found the bugs of course. Too bad it didn't work >then. Could have won the tournament. Just analyze the games. > >>-S.
This page took 0.01 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.