RFR (s): 8153755: Improve test: stress/gc/TestStressRSetCoarsening.java

Dmitry Fazunenko dmitry.fazunenko at oracle.com
Tue Apr 12 13:32:00 UTC 2016


Kirill,

Thank you for your notices!
New version
http://cr.openjdk.java.net/~dfazunen/8153755/webrev.02/

Please find some comments inline:

On 12.04.2016 15:14, Kirill Zhaldybin wrote:
> Dmitry,
>
> Just a couple of remarks:
> 1. You could get a number of free regions with WhiteBox method:
> regionCount = (int) WB.g1NumFreeRegions();
>
> It should be more reliable (and simpler) than
>         // How many free regions
>         Runtime rt = Runtime.getRuntime();
>         long used = rt.totalMemory() - rt.freeMemory();
>         long totalFree = rt.maxMemory() - used;
>         regionCount = (int) ((totalFree / regionSize) * 
> heapFractionToAllocate);

Yes, I modified as you suggested.
I kept used/totalFree to print information about system.

>
> 2. You could calculate actual Object[N] size without loop - see 
> gc/testlibrary/Helpers.java detectByteArrayAllocationOverhead as example.

The Helper class seems to do what's needed, but it deals for byte[], 
when I work with Object[]. So, to use the Helper I will have to write an 
extra method.
I prefer to keep my implementation as is, which is relatively simple.

Thanks,
Dima

>
> Thank you.
>
> Regards, Kirill
>
>
> On 11.04.2016 16:09, Dmitry Fazunenko wrote:
>> Hello,
>>
>> May I have a couple of reviews for a small change to the stress test:
>>
>> 1) change logging options (-XX:+PrintGC --> -Xlog:gc*)
>> 2) Use WhiteBox API to obtain size of new Object[N] object
>>
>>
>> https://bugs.openjdk.java.net/browse/JDK-8153755
>> http://cr.openjdk.java.net/~dfazunen/8153755/webrev.00/
>>
>> Testing: locally + RBT
>>
>> Thanks,
>> Dima
>>
>>
>>
>



More information about the hotspot-gc-dev mailing list