| View previous topic :: View next topic |
| Author |
Message |
Vincent Diepeveen
Joined: 09 Mar 2006 Posts: 1738 Location: The Netherlands
|
Post subject: Re: YBWC: Active Reparenting Posted: Wed Apr 11, 2012 10:09 am |
|
|
| mcostalba wrote: |
| diep wrote: |
In other words the situation that there is a position where we might be able to add a cpu occurs basically if in a position we already have a bunch of cores search in parallel AND there is more moves to search in that position Q left.
|
Yes, this is the condition I am looking for in my patch. And I also limit the search to the 'oldest' split points that it means, for each thread, master of a set of split points, I check only the split point highest in the tree so to possibly reparent near the root: this should help getting a longer average life time for the slave. |
Marco that's what i have been trying to do some years ago in Diep as well.
Though not in a master slave model - as diep doesn't have master/slave model, entire search is in shared memory.
What i did do is each cpu keeps track of positions where one is allowed to split. other cpu's also having access to the shared ram, could read this if they wanted to, then send a request to the split owner to split it in at that spot that it found itself. splitowner received request and then splitted in the cpu requesting to split at that spot. Idea was to always split in closest to the root.
That's what i did do.
Took me a few years to realize we can already mathematically prove this to work crap as basically our parallel search already runs genius at the moment that we can split in there.
Because what basically needs to happen for this to work is that cpu's that already tried to split in others, there were no others to split in, as all cpu's were already carrying out a parallel search.
Only AFTER all cpu's already are doing a parallel search and machine busy 100% with 0 cpu's not having a job, then you could come along a position where you can't split as no cpu's are available.
Basically it's a DTS derived idea.
So
a) your idea is not new i already tried it and DTS is based upon it
In old Diep code i called this : HelpYourSelf()
b) it's a lot of work to prove it correct and has overhead
c) it only can give splitpoints to cpu's when all cpu's in the first place already are searching perfectly parallel and nothing is idle. So your machine already scales genius by then.
If something already scales well no need to add this feature.
2 guys already tried. One based its entire SMP upon it (Bob) and another (me) forgot that problem C.
p.s. note that i had turned off the feature quickly as in case of Diep i had not done B - it wasn't bugfree. Hadn't proven it on paper back then... ... and as in case of diep ever cpu is the same suddenly introducing splitowners was a bad idea at all.
p.s.2 one of problems i encountered is that in diep all cpu's are the same, so when one cpu would call an AbortMe() basically all splitpoints would be lost, so i also designed a mechanismback then to have other cpu take over those potential splitpoints |
|
| Back to top |
|
 |
|
| Subject |
Author |
Date/Time |
YBWC: Active Reparenting |
Marco Costalba |
Tue Apr 10, 2012 5:38 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Tue Apr 10, 2012 6:23 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Tue Apr 10, 2012 7:11 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Tue Apr 10, 2012 7:54 pm |
Re: YBWC: Active Reparenting |
Marco Costalba |
Tue Apr 10, 2012 8:48 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Wed Apr 11, 2012 10:09 am |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Mon Apr 16, 2012 6:53 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Tue Apr 10, 2012 11:11 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Tue Apr 10, 2012 11:33 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Mon Apr 16, 2012 7:07 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Mon Apr 16, 2012 8:55 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Tue Apr 17, 2012 2:54 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Tue Apr 17, 2012 5:42 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Tue Apr 17, 2012 8:01 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Tue Apr 17, 2012 9:34 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Tue Apr 17, 2012 9:46 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Tue Apr 17, 2012 10:30 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Tue Apr 17, 2012 11:43 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Wed Apr 18, 2012 1:05 am |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Wed Apr 18, 2012 11:24 am |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Wed Apr 18, 2012 1:01 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Wed Apr 18, 2012 7:34 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Wed Apr 18, 2012 11:18 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Thu Apr 19, 2012 7:01 am |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Thu Apr 19, 2012 12:46 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Thu Apr 19, 2012 2:40 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Thu Apr 19, 2012 9:32 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Thu Apr 19, 2012 7:08 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Thu Apr 19, 2012 9:37 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Thu Apr 19, 2012 9:48 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Thu Apr 19, 2012 6:59 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Thu Apr 19, 2012 5:30 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Tue Apr 17, 2012 11:29 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Wed Apr 18, 2012 12:58 am |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Tue Apr 17, 2012 11:18 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Tue Apr 17, 2012 5:37 pm |
Re: YBWC: Active Reparenting |
Álvaro Begué |
Tue Apr 10, 2012 6:27 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Tue Apr 10, 2012 6:44 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Tue Apr 10, 2012 9:39 pm |
Re: YBWC: Active Reparenting |
Marco Costalba |
Wed Apr 11, 2012 6:06 am |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Thu Apr 12, 2012 1:20 am |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Sun Apr 15, 2012 10:04 am |
Re: YBWC: Active Reparenting |
Marco Costalba |
Sun Apr 15, 2012 10:22 am |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Sun Apr 15, 2012 2:39 pm |
Re: YBWC: Active Reparenting |
Marco Costalba |
Mon Apr 16, 2012 5:29 am |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Mon Apr 16, 2012 5:54 pm |
Re: YBWC: Active Reparenting |
Marco Costalba |
Tue Apr 17, 2012 11:30 am |
Re: YBWC: Active Reparenting |
Marco Costalba |
Wed Apr 18, 2012 6:31 am |
Re: YBWC: Active Reparenting |
Rein Halbersma |
Wed Apr 11, 2012 7:23 am |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Wed Apr 11, 2012 9:07 am |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Wed Apr 11, 2012 9:31 am |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Wed Apr 11, 2012 12:14 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Wed Apr 11, 2012 1:54 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Thu Apr 12, 2012 1:28 am |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Tue Apr 17, 2012 5:49 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Tue Apr 17, 2012 11:10 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Wed Apr 18, 2012 12:50 am |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Wed Apr 18, 2012 7:38 pm |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Wed Apr 18, 2012 11:30 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Thu Apr 19, 2012 7:08 am |
Re: YBWC: Active Reparenting |
Daniel Shawul |
Thu Apr 19, 2012 12:15 pm |
Re: YBWC: Active Reparenting |
Vincent Diepeveen |
Thu Apr 19, 2012 1:53 pm |
Re: YBWC: Active Reparenting |
Robert Hyatt |
Thu Apr 19, 2012 7:11 pm |
Re: GPUs |
Srdja Matovic |
Mon Jun 04, 2012 5:08 pm |
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
|
|