RFR: 8241670: Enhance heap region size ergonomics to improve OOTB performance

[Stefan Johansson]

Thanks for the reviews,

Will update per your suggestions before pushing.

Cheers,
Stefan

On 2020-03-31 12:49, Thomas Schatzl wrote:
> Hi,
> 
> On 31.03.20 11:42, Stefan Johansson wrote:
>> Hi,
>>
>> Please review this enhancement to improve the out of the box 
>> performance of G1.
>>
>> Webrev: http://cr.openjdk.java.net/~sjohanss/8241670/00/index.html
>> JBS: https://bugs.openjdk.java.net/browse/JDK-8241670
>>
>> Summary
>> The default heap region size determined at startup used the initial 
>> and max heap size to calculate a region size so that the heap would 
>> have at least 2048 regions (if possible). This proposed patch will 
>> change this to:
>> 1) Only consider Max to make it easy to explain and avoid strange 
>> situations where -Xms or the lack of it will cause different region 
>> size for the same max heap size.
>> 2) Round up the region size to next power of 2, since we've seen many 
>> cases where a larger region size is beneficial.
>> 3) Keep the 2048 target for now since the other two changes, will have 
>> good effect on choosing a larger region size for heaps above 2G.
>>
>> Testing
>> Mach5 tier1-4, aurora performance run for sanity and manual 
>> performance testing to verify results.
>>
> 
> - heapRegion.cpp: s/benificial/beneficial
> 
> - heapRegionBounds.hpp: Maybe remove the "(based on the max heap size)." 
> comment part. Apparently we forgot to update that last time, so we 
> probably will next time too. The code is simple enough too.
> 
> - maybe update copyrights while you are at it.
> 
> No need to re-review these comment updates.
> 
> Looks good.
> 
> Thanks,
>    Thomas