[jira] [Commented] (TAJO-1058) Implement off-heap hash table for hash aggregation

["Hyunsik Choi (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/TAJO-1058?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14159310#comment-14159310 ] 

Hyunsik Choi commented on TAJO-1058:
------------------------------------

Off-heap hash table should keep aggregation contexts into off-heap memory space. For this, it requires a different kind function implementation. TAJO-1092 will resolve this problem.

> Implement off-heap hash table for hash aggregation
> --------------------------------------------------
>
>                 Key: TAJO-1058
>                 URL: https://issues.apache.org/jira/browse/TAJO-1058
>             Project: Tajo
>          Issue Type: Sub-task
>          Components: physical operator
>            Reporter: Hyunsik Choi
>            Assignee: Hyunsik Choi
>              Labels: block_iteration
>             Fix For: block_iteration
>
>
> See the title.
> Currently, we use HashMap for hash aggregation. It causes GC overheads when the number of distinct hash keys is over tens of millions. We need to implement an alternative to the current implementation. This implementation should consider block iteration addressed in TAJO-1041.



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