[
https://issues.apache.org/jira/browse/AXIS2-4878?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dan Armstrong updated AXIS2-4878:
---------------------------------
Description:
We return arrays of complex objects - all of the same type for any particular
web services call. Axis2 was unusably slow. Upon closer inspection, it was
spending much of its time in BeanUtil. The #1 problem is that it is performing
introspection repeatedly for each and every object. Introspection is very
expensive - calling the methods once you find them is cheap.
The solution is to perform the introspection and cache the results. I added a
few lines of code to BeanUtil.java and brought our request from 48 seconds down
to 8. This solution may not be exactly how you would implement it, but the
idea is sound and the results are very impressive:
import java.util.concurrent.*;
private static final ConcurrentMap<Class<?>,BeanInfo> beanInfoCache = new
ConcurrentHashMap<Class<?>,BeanInfo>();
// About line 140, where calls Introspector.getBeanInfo:
BeanInfo beanInfo = beanInfoCache.get(beanClass);
if(beanInfo==null) beanInfoCache.put(beanClass, beanInfo =
Introspector.getBeanInfo(beanClass, beanClass.getSuperclass()));
The map would probably need to use weak references to not prevent class
unloading. And you may not want to depend on the Java 5 generics and
concurrent collections. But the point remains - without this POJO is too slow
(rhyme intended).
Please add this to the Axis2 to benefit others.
Thank you,
Dan Armstrong
AO Industries, Inc.
was:
We return arrays of complex objects - all of the same type for any particular
web services call. Axis2 was unusably slow. Upon closer inspection, it was
spending much of its time in BeanUtil. The #1 problem is that it is performing
introspection repeatedly for each and every object. Introspection is very
expensive - calling the methods once you find them is cheap.
The solution is to perform the introspection and cache the results. I added a
few lines of code to BeanUtil.java and brought our request from 48 seconds down
to 8. This solution may not be exactly how you would implement it, but the
idea is sound and the results are very impressive:
import java.util.concurrent.*;
private static final ConcurrentMap<Class<?>,BeanInfo> beanInfoCache = new
ConcurrentHashMap<Class<?>,BeanInfo>();
// About line 140, where calls Introspector.getBeanInfo:
BeanInfo beanInfo = beanInfoCache.get(beanClass);
if(beanInfo==null) beanInfoCache.put(beanClass, beanInfo =
Introspector.getBeanInfo(beanClass, beanClass.getSuperclass()));
The map would probably need to use weak references to not prevent class
unloading. And you may not want to depend on the Java 5 generics and
concurrent collections. But the point remains - without this POJO to too slow
(rhyme intended).
Please add this to the Axis2 to benefit others.
Thank you,
Dan Armstrong
AO Industries, Inc.
> Simple BeanInfo cache in BeanUtil.java helped as go from 48 second to 8
> second request
> --------------------------------------------------------------------------------------
>
> Key: AXIS2-4878
> URL: https://issues.apache.org/jira/browse/AXIS2-4878
> Project: Axis2
> Issue Type: Improvement
> Components: adb
> Affects Versions: 1.5.2
> Environment: Linux (Debian Lenny x86_64), JDK 1.6.0, Tomcat 6 (in
> NetBeans 6.5.1). Deployed POJO service as .aar file.
> Reporter: Dan Armstrong
> Original Estimate: 0.25h
> Remaining Estimate: 0.25h
>
> We return arrays of complex objects - all of the same type for any particular
> web services call. Axis2 was unusably slow. Upon closer inspection, it was
> spending much of its time in BeanUtil. The #1 problem is that it is
> performing introspection repeatedly for each and every object. Introspection
> is very expensive - calling the methods once you find them is cheap.
> The solution is to perform the introspection and cache the results. I added
> a few lines of code to BeanUtil.java and brought our request from 48 seconds
> down to 8. This solution may not be exactly how you would implement it, but
> the idea is sound and the results are very impressive:
> import java.util.concurrent.*;
> private static final ConcurrentMap<Class<?>,BeanInfo> beanInfoCache = new
> ConcurrentHashMap<Class<?>,BeanInfo>();
> // About line 140, where calls Introspector.getBeanInfo:
> BeanInfo beanInfo = beanInfoCache.get(beanClass);
> if(beanInfo==null) beanInfoCache.put(beanClass, beanInfo =
> Introspector.getBeanInfo(beanClass, beanClass.getSuperclass()));
> The map would probably need to use weak references to not prevent class
> unloading. And you may not want to depend on the Java 5 generics and
> concurrent collections. But the point remains - without this POJO is too
> slow (rhyme intended).
> Please add this to the Axis2 to benefit others.
> Thank you,
> Dan Armstrong
> AO Industries, Inc.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
