[jira] [Commented] (HBASE-6913) Implement new data block encoding algorithm that combines the advantages of FAST_DIFF and DIFF_KEY

["ramkrishna.s.vasudevan (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HBASE-6913?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14211948#comment-14211948 ] 

ramkrishna.s.vasudevan commented on HBASE-6913:
-----------------------------------------------

Currently in FAST_DIFF we don't repeat the values if they are exactly same.  But we are not trying to write the part of the value that is not repeating and just indicating the common part that is repeating as we do in the key part.

But doing this will have a problem - we will lose the optimization done in HBASE-10801 where we currently don't copy the value part when the KVs are taken up stream for comparison during seek or during fetching a KV to be sent to the client.
Once we start encoding the value part also then we may have to copy the value also before we move on to the next KV.

> Implement new data block encoding algorithm that combines the advantages of FAST_DIFF and DIFF_KEY
> --------------------------------------------------------------------------------------------------
>
>                 Key: HBASE-6913
>                 URL: https://issues.apache.org/jira/browse/HBASE-6913
>             Project: HBase
>          Issue Type: Improvement
>            Reporter: Mikhail Bautin
>            Assignee: Mikhail Bautin
>
> We have noticed that both FAST_DIFF and DIFF_KEY encoding algorithms have some drawbacks in that they don't take advantage of certain types of redundancies in keys/values. We need to implement a new algorithm that combines the most useful properties of these two algorithms, and specifically unit-test that various types of redundancies are removed.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)