RFR: 8155087: Calcualtion in other_time_ms() is incorrect

Stefan Johansson stefan.johansson at oracle.com
Wed Apr 27 09:22:15 UTC 2016


Thanks for reviewing Thomas,

On 2016-04-26 17:00, Thomas Schatzl wrote:
> Hi,
>
> On Tue, 2016-04-26 at 14:58 +0200, Stefan Johansson wrote:
>> Hi,
>>
>> Please review this small fix for:
>> https://bugs.openjdk.java.net/browse/JDK-8155087
>>
>> Webrev:
>> http://cr.openjdk.java.net/~sjohanss/8155087/hotspot.00/
>>
>> Summary:
>> The calculation in other_time_ms() used the average time of some
>> known parallel phases to calculate the "other" time. In the past this
>> might have been the best way to do it but now we actually measure the
>> parallel time and can use this instead.
> Neither this summary nor the CR actually expands on why this
> calculation is incorrect. Could some note about this be added?
Yes, I was a bit brief in the description. Using the averages and sum 
them is not safe because of the way we calculate these averages. The 
average take into account the number of thread contributing to a single 
phase and the sum might be incorrect if not all thread have contributed 
to all phases.

Thanks,
Stefan

> Looks good otherwise.
>
> Thomas
>



More information about the hotspot-gc-dev mailing list