Hi Kohsuke,
From limited testing this what we have seen so far:
1) if setValidating is set to true see lot of InterLeaveExp objects and the
unmarshall time takes between 2 milliseconds and 70 seconds. The verbose GC
allocation failure is frequent
2) if setValidating is set to false the time taken consistently is less than
5 ms but max we have seen is 213 milliseconds. verbose GC looks lot better
fewer allocation failures
I will update if we see anything new after performance testing.
Regards,
kesavan
>From: Kohsuke Kawaguchi <Kohsuke.Kawaguchi_at_Sun.COM>
>Reply-To: users_at_jaxb.dev.java.net
>To: users_at_jaxb.dev.java.net
>Subject: Re: InterleaveExp object growth
>Date: Tue, 22 Feb 2005 18:05:37 -0800
>
>Kesavan T.S wrote:
>>Hi Kohsuke,
>>I tried the switching to using a single jaxb context and reusing
>>Unmrshaller and upgraded to jwdsp 1.5. But from what I see the
>>InterLeaveExp objects do not show up in heap dump only when I do not
>>call setValidating(true) .
>
>It will show up if you use a validating unmarshaller. But if you use a
>single JAXBContext, you should see a lot less of them than before. Or do
>you still see the same number?
>
>>Also the marshalling time taken is very inconsistent when unmarshalling
>>The time range in the test we have seen to unmarshall iis between 2ms to
>>60 seconds.
>
>Interesting. I remember somebody observed a similar problem --- the
>unmarshalling performance varies greatly, but we haven't identified what's
>causing it. Does that only happen when you do setValidating(true)?
>
> > Does every call to setValidating() have any performance penalty.
>
>It shouldn't be. The first time you do it, the RI computes things at stores
>it on JAXBContext. Then the successive setValidating(true) on any
>unmarshaller that came from the same JAXBContext should complete almost
>instantly.
>
>>Also how does the JAXB choose its XML parser we are using xercesImpl.jar
>>1.4.4 verison. Could that impact the performance.
>>I really your appreciate help and recommendation. Our transaction volumes
>>are increasing I would appreciate any recommendation to improve the
>>performance
>
>We just do SAXParserFactory.newInstance() to get a parser to use. Yes, a
>parser generally affects the performance, and faster parser (like Piccolo)
>does improve a performance.
>
>Also, validation is known to be slow. You should think about turning it
>off.
>
>--
>Kohsuke Kawaguchi
>Sun Microsystems kohsuke.kawaguchi_at_sun.com
>
>---------------------------------------------------------------------
>To unsubscribe, e-mail: users-unsubscribe_at_jaxb.dev.java.net
>For additional commands, e-mail: users-help_at_jaxb.dev.java.net
>