[concurrency-interest] RFR [8011215] optimization of CopyOnWriteArrayList.addIfAbsent()

Ivan Gerasimov ivan.gerasimov at oracle.com
Tue Apr 2 18:45:13 EDT 2013


Thank you, Ulf!

> maybe the old code wins for looong arrays, so there could be a 
> threshold to decide between old and new code:

I've modified the benchmark code to test arrays with 90'000 to 100'000 
elements. (Previously was testing 1 to 100 elements.)
The performance gain turns out to be even more significant.
On my machine tests show that with that many elements the new code runs 
40% faster.

Honestly, I didn't expect that. I thought my code might be a bit slower 
and hoped that not much slower.

Sincerely,
Ivan

>
>
> Am 02.04.2013 23:11, schrieb Ivan Gerasimov:
>>
>>> Thanks for this change.  There is a tradeoff here.  If the element 
>>> is never present, then the older code might be a little faster, 
>>> because we can avoid re-traversing the array.  Otherwise, the new 
>>> code is better.
>>
>> I've done a little testing on my side.
>> I used Integer as an underlying type and set length of the array to 
>> the values from 1 to 100.
>> My code shows a little performance gain - approximately 9%.
>> I understand it may not be there for all cases, but at least for some 
>> cases it is there.
>>
>>> I prefer it your way (I hate unneeded allocation), but the code was 
>>> intentionally written the other way.  Let's hear from Doug...
>>>
>>> Martin
>>>
>>>
>>> On Tue, Apr 2, 2013 at 1:38 PM, Ivan Gerasimov 
>>> <ivan.gerasimov at oracle.com <mailto:ivan.gerasimov at oracle.com>> wrote:
>>>
>>>
>>>>         Please review my proposal for the
>>>>         CopyOnWriteArrayList.addIfAbsent() method optimization.
>>>>
>>>> http://washi.ru.oracle.com/~igerasim/webrevs/8011215/webrev/index.html
>>>> <http://washi.ru.oracle.com/%7Eigerasim/webrevs/8011215/webrev/index.html> 
>>>>
>>>>
>>>>
>>>>     This URL is not readable by external reviewers.
>>>
>>>     The webrev has been copied here:
>>>     http://cr.openjdk.java.net/~coffeys/webrev.8011215.ivan/
>>> <http://cr.openjdk.java.net/%7Ecoffeys/webrev.8011215.ivan/>
>>>
>>>
>>>>     The "master" version of CopyOnWriteArrayList is here:
>>>> http://gee.cs.oswego.edu/cgi-bin/viewcvs.cgi/jsr166/src/main/java/util/concurrent/CopyOnWriteArrayList.java?view=markup 
>>>>
>>>>
>>>     Thanks for the link!
>>>     I see that the code in the master version is identical to the one
>>>     I've been working on.
>>>     So the optimization still could be applied.
>>>
>>>     Sincerely,
>>>     Ivan
>>>
>>>
>>
>>
>
>
>



More information about the Concurrency-interest mailing list