[rvm-research] Count reachable objects in nursery space

[AlexN. <a_n...@rb...>]

Hi everyone,

we are working on a dynamic analysis to find methods which produce object
churn. ([1], [2]) 

this analysis needs to know how much objects are reachable at the start and
the end of some profiled method. In [2] it is advised to do this by doing a
garbage collection and counting the objects which are processed. We use the
GenMS collector. The problem here is, if we do the standard collection,
objects from nursery space are copied to mature space at the beginning of
the method. So we have to do a full heap collection to process the moved
objects, too, which is very slow. 

We tried to disable the copy process and didn't released the nursery space.
But this leaded to some problems we cannot solve, because of the complexity
of the GC. 

Is there any easy solution to prevent the collector from moving objects from
nursery space to mature space or is there a easier way to count the
reachable objects in nursery space, independent from normal GC process? 

Regards,
Alex

[1] http://www.cs.berkeley.edu/~aj/cs/jolt.ppt, [2]
http://portal.acm.org/citation.cfm?id=1449955.1449775
-- 
View this message in context: http://www.nabble.com/Count-reachable-objects-in-nursery-space-tp21705106p21705106.html
Sent from the jikesrvm-researchers mailing list archive at Nabble.com.