 | Frequently Asked Questions |
1. Why does search in All Solutions mode take so much time?
Search for checkmate in All Solutions mode (right click on Start button) involves runs in a few heuristical passes (see button 'Show' in Heuristics Selection Dialog) and a pass without heuristics. The final pass without heuristics is still optimized for FINDING mate, not for finding a refutation. This is why Brute Force and Hybrid Brute Force runs may be recommended (see items in menu 'Problem'). They are optimized for finding refutations as fast as possible.
2. Search for Black defence takes very long time, why is it so?
Search for best defence in general is a very time-consuming operation: in case of a #n problem, many #(n-1) problems would have to be solved brute force (or in All Solutions mode) in order to be able to make a precise decision. This is why search for defence by Black for #7+ is heuristical, nonetheless in many cases it may still take long time (particularly if right click is made for more careful analysis). This is why there are cases where manual selection of best defence by Black may be recommended (for example when program doesn't give the result in the expected amount of time)
3. I am using .tmpReg file. Does it have to be reloaded after every program restart?
Yes, temporary data of the problem (hash table built during solving or loaded from file) is removed from memory already during navigation to another problem within .cp file (see Left and Right Arrow buttons in the top of the program window). This is to avoid using too much RAM. This is why after some particularly important long analysis it may be recommended to save this data to a file (menu Advanced->Save temporary register of the current problem). Note also that program has an option of periodical AutoSave of temporary data that may be useful in case of very long analyses. The default values are to save up to 256MB (after LZ4 compression) every 4h of analysis.
4. I selected from menu "Solve and show the whole main variant". J.G.Island solved the main problem quickly but then it started pondering for a very, very long time. Why is it so?
Most probably 'the guilty one' is search for Black defence (see 'Search for Black defence takes very long time, why is it so?') This is why finding the main variant sometimes may involve manual intervention and making a potential best Black's defence manually.
5. J.G.Island found a dual in the 2nd move, displayed both solutions but only the first of these moves was made on the chessboard. How to proceed to see the details of the dual variant?
The move made on the chessboard can be undone (see red circle icon with a white arrow back). Then you can manually make any other move to be able to start analysis for the position after this move.
6. How can I upgrade to a newer version of J.G.Island?
J.G.Island does not have an option of update/upgrade. The new version should be downloaded from jgisland.pl and installed. Installation in the same location/folder is recommended (although if any changes made meanwhile by user in delivered files - like sample_problems.cp - they will be overwritten during installation)
7. What is the best threading selection? Should I allow engine when it tries to force another threading selection than the current setting (4^2)?
Short answer: Although threading is a complex thing, the short answer is: yes, most probably you should let engine force number of threads.
Longer answer: J.G.Island tries to use more threads after some not very intricate analysis, for example if it detects that in the starting position there are more moves that look almost equally strong (for example forcing moves with single defence) than threads. However it has some limitations as far as forcing is concerned, so it can occur in some positions when 4 threads are selected, it tries to force 6 threads, but if you manually select 6 threads, it tries to force to 8 or 9 (admittedly, it may look slightly strange). (Additionally, forcing refers only to the first move, so if 4^2 is selected and 6 threads is forced, then the actual number of main threads is 6 x 4 = 24, not 6^2 = 36)
My experience so far indicates that the maximum limit of (main) threads for relatively efficient processing is total processors (logical and physical) times two (context switching is not very frequent yet in such case). So for a 32-processor machine this limit is around 64 threads. Which further means, that even 4^3 may be not that bad (64 main threads). Of course there are alternatives like 6^2.
Anyway it is also worth noting that in some cases even a single thread can be an optimal selection provided that both conditions below are satisfied:
a) you have a strong suspicion that sorting first two/three white moves is proper (engine finds really the best move) or you force proper moves using Sort Indications or option to Use main variant as sort indications,
b) you have a strong suspicion that initial heuristics selection is the most proper one and heuristics relaxation will not be needed (or you select these heuristics yourself)
Then it is not so reasonable to make the machine busy with as much as 4^3 threads, particularly if you want to use it also for some other purpose meanwhile.
See also hints for efficient usage and useful tricks