[jira] [Commented] (HIVE-7859) Tune zlib compression in ORC to account for the encoding strategy

["Gopal V (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-7859?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14127946#comment-14127946 ] 

Gopal V commented on HIVE-7859:
-------------------------------

Tested this with TPC-H 1000 scale inserts.


|| Type || Time || Size ||
| Snappy | 555.234 seconds |  224.4Gb |
| Zlib speed | 585.137 seconds | 176.7Gb |
| Zlib compression | 767.23 seconds | 154.8Gb |

> Tune zlib compression in ORC to account for the encoding strategy
> -----------------------------------------------------------------
>
>                 Key: HIVE-7859
>                 URL: https://issues.apache.org/jira/browse/HIVE-7859
>             Project: Hive
>          Issue Type: Bug
>          Components: File Formats
>            Reporter: Gopal V
>            Assignee: Gopal V
>         Attachments: HIVE-7859.1.patch, HIVE-7859.2.patch
>
>
> Currently ORC Zlib is slow because several compression strategies ZLib uses is already done by ORC in itself (dictionary, RLE, bit-packing).
> We need to pick between Z_FILTERED, Z_HUFFMAN_ONLY, Z_RLE, Z_FIXED and Z_DEFAULT_STRATEGY according to column stream type.
> For instance an RLE_V2 stream could a use Z_FILTERED compression without invoking the rest of the strategies.
> The string streams can use Z_FIXED compression strategies and so on.
> The core limitation to stick to retain compatibility with the default decompressor, so that these are automatically backward compatible.



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