You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@wink.apache.org by "Bryant Luk (JIRA)" <ji...@apache.org> on 2009/09/09 19:11:57 UTC
[jira] Created: (WINK-175) Change ProvidersRegistry MediaTypeMap to
allow gets to be faster
Change ProvidersRegistry MediaTypeMap to allow gets to be faster
----------------------------------------------------------------
Key: WINK-175
URL: https://issues.apache.org/jira/browse/WINK-175
Project: Wink
Issue Type: Improvement
Components: Common
Affects Versions: 0.2
Reporter: Bryant Luk
Fix For: 0.2
Doug Larson did some more profiling and pointed out that on the ProvidersRegistry MediaTypeMap, using a ConcurrentHashMap over SimpleConcurrentMap actually improves performance by a few percentage points.
{code}
private final SimpleMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>> providersCache =
new SimpleConcurrentMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>>(); ;
{code}
While the cache keys (Class<?>) would not be WeakReferences, the values would still be SoftReferences.
Thoughts?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Closed: (WINK-175) Change ProvidersRegistry MediaTypeMap to
allow gets to be faster
Posted by "Bryant Luk (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/WINK-175?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Bryant Luk closed WINK-175.
---------------------------
> Change ProvidersRegistry MediaTypeMap to allow gets to be faster
> ----------------------------------------------------------------
>
> Key: WINK-175
> URL: https://issues.apache.org/jira/browse/WINK-175
> Project: Wink
> Issue Type: Improvement
> Components: Common
> Affects Versions: 0.2
> Reporter: Bryant Luk
> Assignee: Bryant Luk
> Fix For: 0.2
>
>
> Doug Larson did some more profiling and pointed out that on the ProvidersRegistry MediaTypeMap, using a ConcurrentHashMap over SimpleConcurrentMap actually improves performance by a few percentage points.
> {code}
> private final SimpleMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>> providersCache =
> new SimpleConcurrentMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>>(); ;
> {code}
> While the cache keys (Class<?>) would not be WeakReferences, the values would still be SoftReferences.
> Thoughts?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (WINK-175) Change ProvidersRegistry MediaTypeMap
to allow gets to be faster
Posted by "Michael Elman (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/WINK-175?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12753437#action_12753437 ]
Michael Elman commented on WINK-175:
------------------------------------
If I remember correctly SimpleConcurrentMap doesn't make any cache/soft/weak/references. It simply wraps the provided map with ReentrantReadWriteLock.
It's SoftConcurrentMap that wraps WeakHashMap.
Does Doug Larson makes multi-threaded tests?
SimpleConcurrentMap should have a better performance for multiple threads, since it locks get with a read lock, which allows multiple threads to read from a map simultaneously.
> Change ProvidersRegistry MediaTypeMap to allow gets to be faster
> ----------------------------------------------------------------
>
> Key: WINK-175
> URL: https://issues.apache.org/jira/browse/WINK-175
> Project: Wink
> Issue Type: Improvement
> Components: Common
> Affects Versions: 0.2
> Reporter: Bryant Luk
> Fix For: 0.2
>
>
> Doug Larson did some more profiling and pointed out that on the ProvidersRegistry MediaTypeMap, using a ConcurrentHashMap over SimpleConcurrentMap actually improves performance by a few percentage points.
> {code}
> private final SimpleMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>> providersCache =
> new SimpleConcurrentMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>>(); ;
> {code}
> While the cache keys (Class<?>) would not be WeakReferences, the values would still be SoftReferences.
> Thoughts?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Resolved: (WINK-175) Change ProvidersRegistry MediaTypeMap
to allow gets to be faster
Posted by "Bryant Luk (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/WINK-175?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Bryant Luk resolved WINK-175.
-----------------------------
Resolution: Fixed
Assignee: Bryant Luk
I made the change since the performance looked better.
Out of personal curiosity, I did some quick Googling for when ConcurrentHashMap would lock. One of the better articles I found is: http://www.ibm.com/developerworks/java/library/j-jtp08223/index.html?S_TACT=105AGX02&S_CMP=EDU which explains that the get() operation can often proceed without locking at the Java level. It's an apparently deep understanding of the Java memory model that makes it possible.
> Change ProvidersRegistry MediaTypeMap to allow gets to be faster
> ----------------------------------------------------------------
>
> Key: WINK-175
> URL: https://issues.apache.org/jira/browse/WINK-175
> Project: Wink
> Issue Type: Improvement
> Components: Common
> Affects Versions: 0.2
> Reporter: Bryant Luk
> Assignee: Bryant Luk
> Fix For: 0.2
>
>
> Doug Larson did some more profiling and pointed out that on the ProvidersRegistry MediaTypeMap, using a ConcurrentHashMap over SimpleConcurrentMap actually improves performance by a few percentage points.
> {code}
> private final SimpleMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>> providersCache =
> new SimpleConcurrentMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>>(); ;
> {code}
> While the cache keys (Class<?>) would not be WeakReferences, the values would still be SoftReferences.
> Thoughts?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (WINK-175) Change ProvidersRegistry MediaTypeMap
to allow gets to be faster
Posted by "Bryant Luk (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/WINK-175?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12753607#action_12753607 ]
Bryant Luk commented on WINK-175:
---------------------------------
Yes, Doug has multiple threads running (4-CPU machine and 15 concurrent client threads).
ConcurrentHashMap is able to perform the get()s without needing to acquire a lock. The (what I believe) common usage scenario that we're looking at is an application that reads/writes the same type of data repeatedly.
> Change ProvidersRegistry MediaTypeMap to allow gets to be faster
> ----------------------------------------------------------------
>
> Key: WINK-175
> URL: https://issues.apache.org/jira/browse/WINK-175
> Project: Wink
> Issue Type: Improvement
> Components: Common
> Affects Versions: 0.2
> Reporter: Bryant Luk
> Fix For: 0.2
>
>
> Doug Larson did some more profiling and pointed out that on the ProvidersRegistry MediaTypeMap, using a ConcurrentHashMap over SimpleConcurrentMap actually improves performance by a few percentage points.
> {code}
> private final SimpleMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>> providersCache =
> new SimpleConcurrentMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>>(); ;
> {code}
> While the cache keys (Class<?>) would not be WeakReferences, the values would still be SoftReferences.
> Thoughts?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (WINK-175) Change ProvidersRegistry MediaTypeMap
to allow gets to be faster
Posted by "Hudson (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/WINK-175?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12753948#action_12753948 ]
Hudson commented on WINK-175:
-----------------------------
Integrated in Wink-Trunk-JDK1.5 #145 (See [http://hudson.zones.apache.org/hudson/job/Wink-Trunk-JDK1.5/145/])
Update ProvidersRegistry with ConcurrentHashMap
See []
> Change ProvidersRegistry MediaTypeMap to allow gets to be faster
> ----------------------------------------------------------------
>
> Key: WINK-175
> URL: https://issues.apache.org/jira/browse/WINK-175
> Project: Wink
> Issue Type: Improvement
> Components: Common
> Affects Versions: 0.2
> Reporter: Bryant Luk
> Fix For: 0.2
>
>
> Doug Larson did some more profiling and pointed out that on the ProvidersRegistry MediaTypeMap, using a ConcurrentHashMap over SimpleConcurrentMap actually improves performance by a few percentage points.
> {code}
> private final SimpleMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>> providersCache =
> new SimpleConcurrentMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>>(); ;
> {code}
> While the cache keys (Class<?>) would not be WeakReferences, the values would still be SoftReferences.
> Thoughts?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (WINK-175) Change ProvidersRegistry MediaTypeMap
to allow gets to be faster
Posted by "Michael Elman (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/WINK-175?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12753612#action_12753612 ]
Michael Elman commented on WINK-175:
------------------------------------
AFAIK ConcurrentHashMap uses ReentrantLock and not ReentrantReadWriteLock.
Anyway, if you think that the tests are good and ConcurrentHashMap has better performance over SimpleConcurrentMap, feel free to change it.
> Change ProvidersRegistry MediaTypeMap to allow gets to be faster
> ----------------------------------------------------------------
>
> Key: WINK-175
> URL: https://issues.apache.org/jira/browse/WINK-175
> Project: Wink
> Issue Type: Improvement
> Components: Common
> Affects Versions: 0.2
> Reporter: Bryant Luk
> Fix For: 0.2
>
>
> Doug Larson did some more profiling and pointed out that on the ProvidersRegistry MediaTypeMap, using a ConcurrentHashMap over SimpleConcurrentMap actually improves performance by a few percentage points.
> {code}
> private final SimpleMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>> providersCache =
> new SimpleConcurrentMap<Class<?>, SoftReference<SimpleMap<MediaType, List<ObjectFactory<T>>>>>(); ;
> {code}
> While the cache keys (Class<?>) would not be WeakReferences, the values would still be SoftReferences.
> Thoughts?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.