Currently we run approx. 30-40 games in parallel.

The assimilator and the work_generator are cron jobs which run every few minutes, therefore it shouldn't be a long period without work.

We generate 4 workunits for each halfmove. May be we can reduce it in future to 3. The ordering decides the BOINC standard software.

CPU: 3.37; P:0.009400
CPU: 4.29; P:0.012600
CPU: 5.28; P:0.016000
CPU: 6.25; P:0.019400
CPU: 7.18; P:0.022600
CPU: 8.11; P:0.025800
CPU: 9.03; P:0.029000
CPU: 9.95; P:0.032200

These are some debug info, the CPU time in sec and the percentage of completion (last row: in 9.95 sec 3,22 % of the w/u completed.
____________
Joerg
Chess960athome.org / Chess-960.org

You can see it in your boinc manager: the name of the workunit is chess960_xxx_yyy_z, where xxx is the game number, yyy is the workunit (half move) in the game and z is the number of the user for this workunit (usually 1-4).

Unfortunately the project is out of work right now so I cannot check this... Why is it that it runs out of workunits anyway? Doesn't it advance a game and generate a new move every time (or every third time if the replication number is three) it receives a workunit?

How does the system decide that number z? Are positions with greater branching factor sent to more people or..?

What is the "stderr out" field included in every workunit? Can I get some useful information out of this:

CPU: 3.37; P:0.009400
CPU: 4.29; P:0.012600
CPU: 5.28; P:0.016000
CPU: 6.25; P:0.019400
CPU: 7.18; P:0.022600
CPU: 8.11; P:0.025800
CPU: 9.03; P:0.029000
CPU: 9.95; P:0.032200

____________
Communication usually fails, except by accident.

So is one half-move always one workunit? A thirty-move game takes sixty workunits?

Yes, right.
But in average we have 60-80 moves, means 120-160 workunits per games.

You can see it in your boinc manager: the name of the workunit is chess960_xxx_yyy_z, where xxx is the game number, yyy is the workunit (half move) in the game and z is the number of the user for this workunit (usually 1-4).

____________
Joerg
Chess960athome.org / Chess-960.org

So is one half-move always one workunit? A thirty-move game takes sixty workunits?
____________
Communication usually fails, except by accident.

In my opinion the scenario 2 is a lot better. It's a case of a breadth-first search as opposed to the first scenario, a depth-first approach.

We use scenario 2 in the project.

if the aim of this project is to develop an opening book database for Chess960, then I think there's no point in playing complete games if we just have a good evaluation algorithm, right? (At least at first - the opening book would be expanded later, possibly first on the interesting openings only. Some heuristics would have to be developed to fresh out the 'interesting' ones).

The aim of this project is not in first priority to develop an opening book.
We want to create a (big) collection of games to publish it on the internet for public use. Of course it is also possible to develop a book from it but it's not our focus.

The information about this project is somewhat cluttered around the site and the forums and I was a bit disappointed when there was no new post explaining the current state of the project when the account creation was re-enabled.

We played arround 50-60 games so far, that means we are at the very beginning. We still have some smaller problems and unsolved tasks. For it we search for volunteers with experience in chess engine programming.

To see the current status, take a look at the project information and visit the database on chess-960.org to re-play the games.

I welcome especially chess players (on each level) to take part on this project.




____________
Joerg
Chess960athome.org / Chess-960.org

Hi,

I've been monitoring the site and just joined in immediately when the account creation became available. :)

At this thread there were two possibilities of how to go on about the project:

1. play a complete game with two egines with a command line chess interface to control the egines
In this case one game is one WU.
2. play a single move with one engine and use the result to generate the input for the next move.
In this scenario a single move (in sense of chess it's a half move).

In my opinion the scenario 2 is a lot better. It's a case of a breadth-first search as opposed to the first scenario, a depth-first approach.

(I just realized that actually I don't know but I've allowed myself to assume that BFS is the way that chess-engines generally use?)

If the aim of this project is to develop an opening book database for Chess960, then I think there's no point in playing complete games if we just have a good evaluation algorithm, right? (At least at first - the opening book would be expanded later, possibly first on the interesting openings only. Some heuristics would have to be developed to fresh out the 'interesting' ones).

The information about this project is somewhat cluttered around the site and the forums and I was a bit disappointed when there was no new post explaining the current state of the project when the account creation was re-enabled.

What I think should be done is have each workunit include a starting position and one move. Then the task would be to calculate all moves within, say, 10 or 11 ply. These should be reasonable, judging by the fact how fast Chessmaster 9 does the "Solve for mate" command (the first 5 ply in a second or so, the 6th ply in another and the 7th ply in some two minutes on my already a rather old Athlon XP 1700+ / 1 GB). Next plies always take exponentially large time, but until 10 ply I think it goes in total of under 10 hours anyway.

Please excuse me, I don't mean to sound like a know-it-all, I've just been following the conversations here and been urging to ask my questions and maybe even give my opinions, but without an account I've been unable to do so. :/
____________
Communication usually fails, except by accident.