[
https://issues.apache.org/jira/browse/HBASE-15543?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15216272#comment-15216272
]
ramkrishna.s.vasudevan commented on HBASE-15543:
------------------------------------------------
bq.Is the reported 28% creation of new Cell or new Cell reference when the Cell
is being made against a backing BB that is offheap from BucketCache?
I would think this is the case where a Cell reference is made from the
underlying HDFS BB. Because generally in case of compaction we don't read from
the cache? Even if we do then this cell should be formed from the offheap BB
incase of Bucket cache. ( NON DBE case).
If we could directly copy this buffer's contents corresponding to the cell to
the new file writer's buffer it is going to be easier. We can avoid this object
creation.
Ya in case of DBE yes- we will do the copy of the value part and I don think we
can avoid that.
> HFileScannerImpl.getCell() allocates cell object on every call
> --------------------------------------------------------------
>
> Key: HBASE-15543
> URL: https://issues.apache.org/jira/browse/HBASE-15543
> Project: HBase
> Issue Type: Sub-task
> Reporter: Vladimir Rodionov
> Assignee: Vladimir Rodionov
>
> This accounts for 28% of memory allocation in compaction thread.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
