[infinispan-issues] [JBoss JIRA] (ISPN-871) Out-of-heap data container implementation

[Adrian Nistor (JIRA)]

    [ https://issues.jboss.org/browse/ISPN-871?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12746261#comment-12746261 ] 

Adrian Nistor commented on ISPN-871:
------------------------------------

+1. I have a hunch that a pure java solution based on nio buffers can perform better than a native implementation. Not sure how much context switching overhead is added by JNI of modern JVMs but a few years ago it was pretty slow (also depending on the number of method arguments). If you need lots of very granular JNI calls (and DataContainer interface is such a case) and the arguments/return value need complex marshalling (and memory copying) then this becomes relevant.
                
> Out-of-heap data container implementation
> -----------------------------------------
>
>                 Key: ISPN-871
>                 URL: https://issues.jboss.org/browse/ISPN-871
>             Project: Infinispan
>          Issue Type: Feature Request
>          Components: Configuration
>            Reporter: Manik Surtani
>            Assignee: Tristan Tarrant
>              Labels: gc, jni, native, performance, research
>             Fix For: 6.0.0.Final
>
>
> The {{DataContainer}} interface could be implemented using a off-heap impl in C, using a wrapper around [TBB|http://threadingbuildingblocks.org/]'s concurrent hashmap.
> Cheap and easy way, no memory management needed, at worst case same performance as the Java CHM-like impl of the data container + some JNI overhead.  Potential benefit of large data heaps.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira