db-jdo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Craig Russell <Craig.Russ...@Sun.COM>
Subject Re: JIRA JDO-13
Date Tue, 07 Jun 2005 18:33:03 GMT
Hi Michael,

I'd like to defer the performance issues until we have gotten more of  
the tests to run and have a clear idea of the end of the project (get  
all of the tests written for the mapping as well as new JDO 2 features).


On Jun 7, 2005, at 11:06 AM, Michelle Caisse wrote:

> Michael,
> That sounds like a useful experiment.  I think Derby's performance  
> is generally supposed to be good, though.  I don't know how Craig  
> wants to prioritize this relative to our other issues.
> -- Michelle
> Michael Watzek wrote:
>> Hi Michelle,
>> I still use JPOX 1.1.0-beta-3. One TCK run takes about 2 hours. So  
>> I'm confident that the two runs will finish overnight :-).
>> Does it make sense to use MySQL instead of Derby - just to make  
>> sure it's not Derby slowing down the performance? We have to adapt  
>> the schema files, the URL and the driver name in jdori.properties,  
>> and the classpath.
>> Regards,
>> Michael
>> Andy Jefferson wrote:
>>>> Sounds good, thanks.  I found that the 6/6 JPOX build ran 10x  
>>>> slower
>>>> than the 6/3 build -- 10 hours to run the TCK.  I am running the  
>>>> 6/7
>>>> build right now, and I'm not sure what the final time will be.   
>>>> At first
>>>> it seemed to be running pretty quickly, but now that it's doing the
>>>> fieldtypes tests, it is very slow.  So you may want to use an  
>>>> older JPOX
>>>> build if you hope to get both application and datastore identity  
>>>> done in
>>>> one night!
>>> Hi Michelle,
>>> have you got "org.jpox.autoStartMechanism" set to anything in  
>>> particular ? The fact that things slow down is symptomatic of  
>>> JPOX progressively loading up all of the classes that it has ever  
>>> encountered during its lifetime with the DB schema (JPOX_TABLES  
>>> table) that it connects to. So with each test, it finds more  
>>> classes and so loads up the MetaData of the previous ones, plus  
>>> the new ones. If you set this property to "None" it will start  
>>> from scratch each time (and so not go off an load up other  
>>> classes not used in that test). The log would tell you what is  
>>> taking the time
>>> Nothing significant has changed in the period you mention (AFAIK)  
>>> and I see no change in runtime speed here on our unit tests.

Craig Russell
Architect, Sun Java Enterprise System http://java.sun.com/products/jdo
408 276-5638 mailto:Craig.Russell@sun.com
P.S. A good JDO? O, Gasp!

View raw message