[jira] [Updated] (LUCENE-6668) Optimize SortedSet/SortedNumeric storage for the few unique sets use-case

["Adrien Grand (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/LUCENE-6668?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Adrien Grand updated LUCENE-6668:
---------------------------------
    Attachment: LUCENE-6668.patch

Here is a patch: it uses table encoding for SortedSet/SortedNumeric if the sum of the sizes of all unique sets is 256 or less. If my math is correct, this means it will always be used if there are 6 unique values or less (given that the sum of the sizes of all possible subsets would be 192), and might be used if the number of unique values is between 7 and 256.

> Optimize SortedSet/SortedNumeric storage for the few unique sets use-case
> -------------------------------------------------------------------------
>
>                 Key: LUCENE-6668
>                 URL: https://issues.apache.org/jira/browse/LUCENE-6668
>             Project: Lucene - Core
>          Issue Type: Improvement
>            Reporter: Adrien Grand
>            Assignee: Adrien Grand
>            Priority: Minor
>         Attachments: LUCENE-6668.patch
>
>
> Robert suggested this idea: if there are few unique sets of values, we could build a lookup table and then map each doc to an ord in this table, just like we already do for table compression for numerics.
> I think this is especially compelling given that SortedSet/SortedNumeric are our two only doc values types that use O(maxDoc) memory because of the offsets map. When this new strategy is used, memory usage could be bounded to a constant.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org