[gecode-users] Bugs in Gecode/J

[Malcolm Ryan]

Sorry, but that doesn't seem to be it. I've run it in single-processor  
mode on both the Mac and in Linux and the crashes are still occurring.  
It may be that some of the intermittent ones have stopped, but the  
repeatable ones remain.

Malcolm

On 21/04/2008, at 7:34 PM, Guido Tack wrote:

> I'm now certain that it's the concurrent GC that's causing the  
> problems (well, because of some bug in Gecode/J, of course).  You  
> can easily work around it by running java with the -XX:- 
> UseParallelGC option.
>
> Cheers,
> 	Guido
>
> Guido Tack wrote:
>
>> Malcolm Ryan wrote:
>>
>>> I don't mean to hassle you, but I just need to know whether you  
>>> think
>>> you can help me with this problem. I am working towards a close  
>>> paper
>>> deadline and I need to know whether to stick with Gecode/J in the  
>>> hope
>>> that we will be able to sort out these problems, or whether I need  
>>> to
>>> look for another solution.
>>
>> I've run some more experiments, and I can finally reproduce your  
>> bug.  It seems to be a memory management / concurrency issue.  I  
>> could not reproduce the bug on any 1- or 2-core machine, but only  
>> with at least 4 cores.  I suspect Java's concurrent garbage  
>> collector frees some memory that is still in use because we didn't  
>> get the locking right.
>> Did you do all your experiments on multi-core machines (i.e., more  
>> than 2 cores)?  If that's really the problem, you could run your  
>> tests with a limit on the number of CPUs used.  On Linux, you can  
>> achieve this using the taskset command.  On Mac OS 10.5, you can  
>> follow this hint to get control of your processors:
>> http://www.macosxhints.com/article.php?story=20071116183942199
>>
>> I hope this helps.  We'll try to figure out the real cause of the  
>> problem, but concurrency bugs are typically hard to find - it may  
>> take some time.
>>
>> Cheers,
>> 	Guido
>>
>> _______________________________________________
>> Gecode users mailing list
>> users at gecode.org
>> https://www.gecode.org/mailman/listinfo/gecode-users
>