Joker 2.0 memory useage

[Matthias Gemuh]

How does one decrease the memory useage of the Joker engines from 128 MB to 32 MB ?

Pupsi and FireFly (amongst other) also eat heavily, even as UCI engines.

Thanks,
Matthias

[Tony Thomas]

The last version I downloaded was 1106. I can just use 19, 20 or 21 on the commandline under arena to make it use 32, 64 or 128MB of Ram respectively.

[Matthias Gemuh]

Thanks, works fine now.

[hgm]

You have a later Joker version than me?


Sorry about the fact that documentation on Joker is still non-existent.
The only argument it accepts is to set the hash-table size. Any value works, as Joker simply uses 2^arg hash buckets of 64 bytes (containing 7 hash entries each). So 20 causes 1M buckets for 64MB hash (7M entries).

The latest Joker version on my downloads page is now 1.1.08, btw. There also is a version 1.1.11 on my website, but it is not linked to yet, so you would have to type the name in the address field yourself.

I am currently testing 1.1.12, as there is some doubt as about if the earlier Jokers do not cheat. It is conceivable that they purge the hash table in the opponent's time, because the printing of the move occurs before the purge loop. However, the fflush to force the printed data out into the pipe to the GUI occurs after the purge loop, so I guess it is OK, as I do believe output to a pipe is buffered. And I also remember that I had to suppress the purging to prevent Joker from forfeiting every game on time in blitz. This helped, and that would be hard to explain if the purging went on in the opponent's time.

Anyway, to make sure I eliminated the purging alltogether, using an aging mechanism to prevent accumulation of obsolete deep entries in 1.1.12. But I am not sure if that works correctly yet.

[Matthias Gemuh]

Hi Muller,
what I call Joker 2.0 is actually "Joker1107w.exe"

Nice game by Neg !

[Event "Mega2 "]
[Site "ChessGUI"]
[White "Pos"]
[Black "Neg03d"]
[Result "0-1"]

1.e4 Nc6 2.Qh5 g6 3.c3 gxh5 4.Na3 Nf6 5.Nb5 Nxe4 6.d3 Nc5 7.Be2 e5 8.Bxh5 Nxd3
9.Ke2 Nf4 10.Bxf4 exf4 11.Rf1 Qe7 12.Kd2 a6 13.Nxc7 Kd8 14.Nxa8 Qd6 15.Ke2 f6
16.Kf3 Ne5 17.Kxf4 Bh6 18.Kf5 Qe6 19.Ke4 Qxa2 20.Re1 Qxb2 21.Nh3 Qxc3 22.Re2 Qc4
23.Kf5 Nf7 24.Ree1 Nd6 25.Kxf6 Ne4 26.Rxe4 Qxe4 27.Rd1 Qxg2 28.Re1 Qxh3
29.Re5 Qxh2 30.Kf5 Qxh5 31.Kf6 Qf3 32.Rf5 Qxf2 33.Rxf2 Be3 34.Rc2 Rf8 35.Kg7 Rf5
36.Kg8 h5 37.Kg7 Bd4 38.Kg6 Rf6 39.Kg7 Ke8 40.Rxc8 Ke7 41.Kh8 Rf8 42.Kh7 Rxc8
43.Kg6 Rxa8 44.Kg5 Rg8 45.Kxh5 Rh8 46.Kg5 Be3 47.Kg6 Rg8 48.Kh7 Rg4 49.Kh8 Rh4
50.Kg8 Rg4 51.Kh8 Bd4 52.Kh7 Rh4 53.Kg8 Rg4 54.Kh7 Rg7 55.Kh6 Be3 56.Kxg7 b5
57.Kh7 Bd4 58.Kg6 Kd6 59.Kf7 Bc5 60.Kg7 Bd4 61.Kf7 Kc6 62.Ke7 Bc5 63.Ke8 d5
64.Kf7 b4 65.Ke6 d4 66.Kf6 Kd7 67.Kf7 Kd6 68.Kf6 Kc7 69.Ke5 Bd6 70.Ke6 Kc8
71.Kxd6 Kd8 72.Ke6 Ke8 73.Ke5 d3 74.Ke4 d2 75.Ke5 Ke7 76.Kf5 Kd8 77.Kf4 Kd7
78.Ke3 Ke8 79.Kxd2
0-1
{ Game Nr. 3711 : Neg03d wins against Pos by GUI TB adjudication
Finalposition 4k3/8/p7/8/1p6/8/3K4/8 b - -,
StartTime = 22:25:46 on 21.07.2007, GameLength = 79 moves, GameDuration = 00:01:16 }


Best,
Matthias.

[hgm]

Yes, NEG typically strips Pos bare. This is actually a game where Pos last unusually long, and which NEG probably would not have won without adjudication. As NEG has no clue what checkmate is, it is unlikely to win KQK, and needs at least 2 or 3 queens to have a good chance on an accidental checkmate. (It had one in the Chess War Promo against Eden2, in 6 moves. )

Normally NEG has about half its material when Pos is reduced to a bare King, and due to a built-in preference for checking moves, this usually leads to a quick accidental mate. NEG scores ~95% against Pos. Here it was unlucky that its Queen stepped into her own shadow. Perhaps I should implement enemy X-rays in a future version. Another bug is that it needlessly refrains from pawn non-captures, because it does not count the non-capture itself as covering the square, and thus too hastily concludes that the square is more often defended than attacked (and thus unsafe). Even when the number of attackers and defenders is both zero! This is why it did not let Pos chase its pawn to promotion, but allowed it to be captured.