JMC-6324: Values in Method Profiling rule result text not correct when filtering

[Joshua Matsuoka]

ping

On Mon, Jan 28, 2019 at 4:06 PM Joshua Matsuoka <jmatsuok at redhat.com> wrote:

> Hi Marcus,
>
> Here's an updated patch:
> http://cr.openjdk.java.net/~jmatsuoka/JMC-6324/webrev.01/
>
> Cheers,
>
> - Josh
>
> On Mon, Jan 28, 2019 at 8:40 AM Marcus Hirt <marcus at hirt.se> wrote:
>
>> Hi Josh,
>>
>> Looks good! Perhaps the percentage shown for the most interesting
>> candidate should be updated to show the percentage of the actual
>> samples taken? This is since the normalization bars we show as
>> backdrop in some columns is in regards to actual events. I think we
>> only need to drop the relation to the theoretical maximum amount of
>> samples in the user's head once, and only to motivate a lower score
>> than expected.
>>
>> For example:
>> The following methods are the most interesting candidates for code
>> optimization:
>> •se.hirt.jmc.tutorial.hotmethods.HolderOfUniqueValues.countIntersection(HolderOfUniqueValues)
>> (24,1 %) 2018-11-09 14:40:39 – 14:41:09
>> ->
>> The following methods are the most interesting candidates for code
>> optimization:
>> •se.hirt.jmc.tutorial.hotmethods.HolderOfUniqueValues.countIntersection(HolderOfUniqueValues)
>> (98,2 %) 2018-11-09 14:40:39 – 14:41:09
>>
>> or possible more verbosely:
>> The following methods are the most interesting candidates for code
>> optimization:
>> •se.hirt.jmc.tutorial.hotmethods.HolderOfUniqueValues.countIntersection(HolderOfUniqueValues)
>> (98,2 % of the samples) 2018-11-09 14:40:39 – 14:41:09
>>
>> Kind regards,
>> Marcus
>>
>> -----Ursprungligt meddelande-----
>> Från: jmc-dev <jmc-dev-bounces at openjdk.java.net> För Joshua Matsuoka
>> Skickat: den 25 januari 2019 17:46
>> Till: jmc-dev at openjdk.java.net
>> Ämne: RFR: JMC-6324: Values in Method Profiling rule result text not
>> correct when filtering
>>
>> Hi,
>>
>> The following patch adds some extra clarity to the Method Profiling rule.
>>
>> Currently the percentage reported by the Method Profiling rule is the
>> percentage of all possible samples in a given period, rather than the
>> actual number of samples. This can lead to some rather misleading results
>> when for example a method takes up 99% of the actual samples in a window,
>> but only constitutes 25% of the theoretical maximum number of samples.
>>
>> This patch keeps track of both the relation to the maximum, as well as
>> the actual percentage of samples and reports both with the rule results
>> (e.g.
>> Method x had y% of possible samples and z% of actual samples)
>>
>> http://cr.openjdk.java.net/~jmatsuoka/JMC-6324/webrev.00/
>>
>> Thoughts?
>>
>> Cheers,
>>
>> - Josh
>>
>>