So many timeout

[Cyrano]

I have some problem with timeout.

I do some tree search based on a timer. The jitter of this search based on log is around few ms. But even with an internal timeout counter of 800ms, i still have timeout in the game. What is the realistic limit to avoid timeout ? 800ms it's still 20% less time than the rules.

Somebody have an idea about the kind of computer the battle are run ? The optimal algorithme are not the same if the cpu is powerfull or not.

[Xrillo]

Perhaps this is related to your problem

[Cyrano]

In java, i use System.nanoTime(). I try also currentTimeMillis() but the result are the same.

The very strange think is when mybot play against itself on my host, the timeout occur after 4 cyles event with a 100ms deadline. There is something strange !

[Cyrano]

I use both function with a timer at 650ms, but there is still some timeout. It's realy depend on Linux scheduler.

There are far less than 1s of compute time.

[Cyrano]

Even at 500ms, there is some issue.

[Cyrano]

When i look to some match, it seems that the cpu is week or that there is a significant difference in the wall clock time and the cpu time.

[iouri_]

I've answered a similar question in another thread, though it was for C++ and not Java.

My advice, without going into Java implementation details, is:

- Make sure you're using wall clock and not CPU clock.
- Check the time out frequently. I think on my computer I end up checking it about every 1-2 ms, and even that is still kinda risky. Judging by the sentence "4 cyles event with a 100ms deadline", it sounds like you might be checking it not often enough. I'd check the time often within a single cycle, and stop early if the time's up.
- Make sure minimal processing is done after the bot is out of time, e.g. do minor cleanup and output the moves to stdout.

[Cyrano]

I use a tree search that stop at timeout. And i use a 500ms timer.

The timer is double using 2 functions : System.currentTimeMillis() System.nanoTime(), the first reaching the limit is used.