Memory question

[Kempelen]

Is there any programmatly way to get the max amount of memory without using swap file? I mean, I would like to be able my engine could calculate and use the max amount of available "fast" memory. If using swap I think then it could be counterproductive, as it could cause the program to be slower.
thanks

[hgm]

I think specifying how much memory a program can use should be left to the user. If a user specifies more memory than the system can handle, it is his own fault, or perhaps he has a reason for it. Giving the program any say in this is just asking for trouble. Even if it knows how much memory exists, it cannot know with how many programs you intend it to share. Other memory-intensive applications could be started later.

If you want to assist noob users in this, it would be better to let the engine print a warning when it detects excessively slow hash acess. (E.g. if the nps is a factor 100 below normal, and reading out the entire hash table takes more than a reasonable amount of time.)

[diep]

Is there any programmatly way to get the max amount of memory without using swap file? I mean, I would like to be able my engine could calculate and use the max amount of available "fast" memory. If using swap I think then it could be counterproductive, as it could cause the program to be slower.
thanks

Yes there is - however do you want it multithreaded, multiprocessing or single threaded?

[Dan Honeycutt]

I think specifying how much memory a program can use should be left to the user.

Yes and no. The user should set the max but the program should have the option to use less. If the user says use 32 mb hash table and program sees physical memory will only accommodate 16 it should be able to set 16.

Best
Dan H.

[hgm]

I don't agree. For one, it is unreasonable to burden all engines with this. If you want this sort of thing, it should be a GUI task. And it is also pretty pointless. Because when the user makes an allocation error it will in almost all cases be a small one (like a factor 2), because he happens to play more engines at once than he happened to do on the last occasion (e.g. an engine-engine game rather than human-engine. Or two games simultaneously rather than one). And the engine would then happily find the requested size to be smaller than the amount of physical memory, and would not complain.

In addition, when some engines would correct for 'over-asking', and others don't, you might not immediately notice when you over-ask in a tourney (when the engines in the first game silently reduced their memory usage), and discover at the end that all other games have been crap. It would be much better for the user if he would have seen immediately that the hard disk goes berserk with swapping, and no depth was reached.

[syzygy]

We're not talking about "all engines".

Fermin simply wants his engine to do something, and his question is simply if anyone can give him a hint on how to achieve this. This has little to do with what others think that engines "should" do. Fermin is not writing a protocol, and he is not laying out rules for others to follow in any other way.

[hgm]

One should never shy of the attempt to convince people of their erroneous ways...

This is a very counter-productive idea, that will almost never do any good to Fermin's engine, and will cause a lot of harm to those using it.

[Edmund]

One should never shy of the attempt to convince people of their erroneous ways...

This is a very counter-productive idea, that will almost never do any good to Fermin's engine, and will cause a lot of harm to those using it.

The optimal TT size is not uniform across all engines for a given time control. As such it should be left to the engines to adjust this size. Rather implement standardized ways for an engine to report hash-table useage (a) allocated space in % of total provided space (b) filled space in % of allocated space. These can be used as indicators and triggers for warnings within the GUI.

[Dan Honeycutt]

I don't agree. For one, it is unreasonable to burden all engines with this. If you want this sort of thing, it should be a GUI task. And it is also pretty pointless. Because when the user makes an allocation error it will in almost all cases be a small one (like a factor 2), because he happens to play more engines at once than he happened to do on the last occasion (e.g. an engine-engine game rather than human-engine. Or two games simultaneously rather than one). And the engine would then happily find the requested size to be smaller than the amount of physical memory, and would not complain.

In addition, when some engines would correct for 'over-asking', and others don't, you might not immediately notice when you over-ask in a tourney (when the engines in the first game silently reduced their memory usage), and discover at the end that all other games have been crap. It would be much better for the user if he would have seen immediately that the hard disk goes berserk with swapping, and no depth was reached.

It's not a burden. As Ron said, it's not mandatory, its an option the engine can exercise if it chooses.

Do you warn your children that the stove is hot or do you let them find out for themselves. I do concede that the latter can be advantageous as far as teaching the lesson.

Best
Dan H.

[hgm]

I have no children, but if I had, I would warn them that the stove is hot. Of course I expect them to touch it anyway, and that will teach them to take my warnings seriously. A priceless lesson...