31 травня 2012 о 22:34 +0200 Eugeniy Meshcheryakov написав(-ла):
> > Machine is Gumstix Overo (OMAP3) with 256MB ram and 200MB swap,
> > 16GB filesystem on microsd card.
> I think SIGKILL is sent by the kernel because of memory exhaustion.
> 
> % /usr/bin/time spark
> ....
> 0.99user 0.39system 0:01.55elapsed 89%CPU (0avgtext+0avgdata 
> 1669200maxresident)k
> 0inputs+0outputs (0major+104408minor)pagefaults 0swaps
> 
> It seems that spark users too large static buffers somewhere (it cannot
> use dynamic memory allocation because it is writen mostly in SPARK).
> 
> Cold you try to run spark with a large swap file, just to see if this fixes
> it?

The largest data structure in the examiner is
contextmanager__ops__file_heap. Setting ContextManagerMaxFiles and 
ContextManagerMaxUnits
in examinerconstants.aps to 1024 reduces memory consumption. I think the
other posibility is to tune the values dependant on ExaminerSize in this
file. The question is what values to use and on which architectures...

Attachment: signature.asc
Description: Digital signature

Reply via email to