[
https://issues.apache.org/jira/browse/ACCUMULO-1409?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13659906#comment-13659906
]
Keith Turner commented on ACCUMULO-1409:
----------------------------------------
bq. At what size does the having the native map truly affect performance?
Its used to sort incoming data. Sorting data w/ C++ is faster than Java. A
difference would be observed if you started writing to an empty Accumulo
instance w/ native maps.
bq. At what size does the having the native map truly affect performance?
The performance of native maps should be invariant with data size. However
sorting incoming data is only one part of what Accumulo does. Accumulo
eventually has to do compactions, which have nothing to do with the native map.
So I suspect you would see the largest performance increase from native maps
when Accumulo is empty. I am thinking Amdahl's law kicks in as more data is
written, compactions start and now native maps are only one part of the system
thats processing data. It would be interesting to run experiments w/ and w/o
native maps for 10M, 100M, 1B, and 10B entries.
> Make MiniAccumuloCluster support native maps
> --------------------------------------------
>
> Key: ACCUMULO-1409
> URL: https://issues.apache.org/jira/browse/ACCUMULO-1409
> Project: Accumulo
> Issue Type: Sub-task
> Reporter: Keith Turner
> Assignee: Corey J. Nolet
> Labels: proposal
> Fix For: 1.6.0
>
>
> If a user sets up a single node accumulo instance and uses it for testing,
> then w/o native maps the performance will not be as good as it could. I did
> some testing of the new GC in java 7, along the way I noticed a significant
> difference between the native map and java map.
> http://kaccumulo.blogspot.com/2012/11/testing-accumulo-with-g1.html
--
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
