hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Hudson (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-5139) Compute (weighted) median using AggregateProtocol
Date Tue, 24 Jan 2012 06:35:42 GMT

    [ https://issues.apache.org/jira/browse/HBASE-5139?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13191941#comment-13191941
] 

Hudson commented on HBASE-5139:
-------------------------------

Integrated in HBase-TRUNK-security #86 (See [https://builds.apache.org/job/HBase-TRUNK-security/86/])
    HBASE-5139 Addendum handles startRow being null for the case where median is in the first
region

tedyu : 
Files : 
* /hbase/trunk/src/main/java/org/apache/hadoop/hbase/client/coprocessor/AggregationClient.java

                
> Compute (weighted) median using AggregateProtocol
> -------------------------------------------------
>
>                 Key: HBASE-5139
>                 URL: https://issues.apache.org/jira/browse/HBASE-5139
>             Project: HBase
>          Issue Type: Sub-task
>            Reporter: Zhihong Yu
>            Assignee: Zhihong Yu
>         Attachments: 5139-v2.txt, 5139.addendum
>
>
> Suppose cf:cq1 stores numeric values and optionally cf:cq2 stores weights. This task
finds out the median value among the values of cf:cq1 (See http://www.stat.ucl.ac.be/ISdidactique/Rhelp/library/R.basic/html/weighted.median.html)
> This can be done in two passes.
> The first pass utilizes AggregateProtocol where the following tuple is returned from
each region:
> (partial-sum-of-values, partial-sum-of-weights)
> The start rowkey (supplied by coprocessor framework) would be used to sort the tuples.
This way we can determine which region (called R) contains the (weighted) median. partial-sum-of-weights
can be 0 if unweighted median is sought
> The second pass involves scanning the table, beginning with startrow of region R and
computing partial (weighted) sum until the threshold of S/2 is crossed. The (weighted) median
is returned.
> However, this approach wouldn't work if there is mutation in the underlying table between
pass one and pass two.
> In that case, sequential scanning seems to be the solution which is slower than the above
approach.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message