Question on GCJ/Boehm Memory Utilization, Part II

Martin Egholm Nielsen martin@egholm-nielsen.dk
Fri Feb 24 08:58:00 GMT 2006


> Craig Vanderborgh wrote:
(And from a new mail, I noticed :-) )

>> We are already running with GC_free_space_divisor set to 20 in our GCJ 
Oh! Didn't notice...

>> build, as you suggested this the last time we were whacked by one of 
>> these memory allocation problems.  But I am curious - can we expect 
>> unpleasant side-effects if we have this parameter set to the 
>> "extreme"?  What do you thing about this?  Is this parameter a subject 
>> for a tuning study, or are there some rules of thumb that can be used 
>> to guesstimate the best value?

> The higher the divisor, the more often GC will run resulting in less 
> real work being done.

But you can always decrease/increase the value at runtime (requires some 
CNI/JNI code instructing the GC) to take care of situations where 
necessary...
Don't know any rules of thumb - guess it's more or less application and 
hw specific (how much memory available, size of allocations, number of 
repeatedly dropped blocks, nature of allocation, etc.)

// Martin



More information about the Java mailing list