<i18n dev> <i18n> Performance patch for DateFormatSymbols.getZoneIndex(String)

[Deven You]

I think Yoshito's suggestion make sense, since the getZoneIndex is an 
internal method, if there is no manual setting to change the timezone, 
The timezone ID won't be changed for one instance of SimpleDateFormt.

I updated the patch webrev[1] for this suggestion and test the 
GetZoneIndexTest.java, the performance is almost the same with webrev.00.

Please review the updated patch[1].

Thanks a lot!

[1] http://cr.openjdk.java.net/~littlee/OJDK-548/webrev.01 
<http://cr.openjdk.java.net/%7Elittlee/OJDK-548/webrev.01>

On 04/11/2012 04:00 AM, Yoshito Umaoka wrote:
> I'm wondering if caching index matched last time (per 
> DateFormatSymbols instance) is sufficient, instead of keeping multiple 
> results in a hash map.
> I guess majority of use cases repeatedly refer the index of the zone 
> used by SimpleDateFormat.
>
> -Yoshito
>
> On 4/9/2012 10:46 AM, Masayoshi Okutsu wrote:
>> Hi Deven,
>>
>> Here are my comments on the proposed changes.
>>
>> - zoneIndexCache should be an instance field because WeakHashMap 
>> isn't thread-safe and the order of IDs in zoneStrings differs in each 
>> DateFormatSymbols.
>>
>> - equals shouldn't be replaced with equalsIgnoreCase because time 
>> zone IDs are (supposed to be) case-sensitive.
>>
>> - Utilize auto-boxing/unboxing.
>>
>> Thanks,
>> Masayoshi
>>
>> On 4/9/2012 12:10 PM, Deven You wrote:
>>> Hi i18n-devs,
>>>
>>> The getZoneIndex() method is expensive and it's performance depends 
>>> on which timezone you're in. This is because the code steps through 
>>> a list of timezones until it finds the current one. This is 
>>> happening over and over again. An idea would be to either cache it 
>>> or rewrite the way we store time zone ids, such as a hashmap instead 
>>> of an array.
>>> This patch[1] is for the cache option.
>>>
>>> Applications which  format/parse dates using SimpleDateFormat 
>>> repeatedly may obtain benefits from this patch, especially when run 
>>> in a timezone far down the zoneStrings array the improvements will 
>>> be even bigger.
>>>
>>> I have written a test case[2] which shows when a timezone which is 
>>> in lower end of the zone strings array, the performance improvement 
>>> this patch can get.
>>>
>>> test results:
>>>
>>> no patch:
>>> The total time is 742 ms!
>>>
>>> with this patch[1] :
>>> The total time is 508 ms!
>>>
>>> If increase the loop times to 1000000, the results are:
>>>
>>> no patch:
>>> The total time is 4743 ms!
>>>
>>> with this patch[1] :
>>> The total time is 2126 ms!
>>>
>>> The java version is:
>>>
>>> /home/deven/hgrps/jdk8/build/linux-i586/j2sdk-image/bin/java -version
>>> openjdk version "1.8.0-internal"
>>> OpenJDK Runtime Environment (build 
>>> 1.8.0-internal-deven_2012_03_14_16_14-b00)
>>> OpenJDK Server VM (build 24.0-b02, mixed mode)
>>>
>>> [1] 
>>> http://cr.openjdk.java.net/~youdwei/DateFormatSymbol_perf/webrev.00/ 
>>> <http://cr.openjdk.java.net/%7Eyoudwei/DateFormatSymbol_perf/webrev.00/> 
>>>
>>> [2] GetZoneIndexTest.java
>


-- 
Best Regards,

Deven

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/i18n-dev/attachments/20120411/7698ab51/attachment.html