[
https://issues.apache.org/jira/browse/OAK-2853?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14644310#comment-14644310
]
Chetan Mehrotra edited comment on OAK-2853 at 7/28/15 1:06 PM:
---------------------------------------------------------------
Updated the benchmark to enable CopyOnRead. With that following numbers are seen
{noformat}
# FullTextSearchTest C min 10% 50% 90% max
N
Oak-Tar 1 67 69 73 98 203
749 //CopyOnRead enabled
Oak-Tar 1 1397 1413 1480 1751 1908
40 //CopyOnRead disabled
Oak-Tar 1 476 485 508 575 663
115 //CopyOnRead enabled, Compression On
{noformat}
After revisiting the numbers it appears that enabling compression adds
overhead. Things might improve with Lucene 5.x (UCENE-5914). This is observed
by
[others|http://stegard.net/2015/05/performance-of-stored-field-compression-in-lucene-4-1/]
also.
So looks like it does not make sense to yet enable compression by default.
/cc [~teofili] [~alex.parvulescu] [~tmueller]. Unless you think otherwise I
would close this issue as WontFix let the status quo be continued
was (Author: chetanm):
Updated the benchmark to enable CopyOnRead. With that following numbers are seen
{noformat}
Oak-Tar 1 67 69 73 98 203
749 //CopyOnRead enabled
Oak-Tar 1 1397 1413 1480 1751 1908
40 //CopyOnRead disabled
{noformat}
> Use default codec for fulltext index
> ------------------------------------
>
> Key: OAK-2853
> URL: https://issues.apache.org/jira/browse/OAK-2853
> Project: Jackrabbit Oak
> Issue Type: Improvement
> Components: lucene
> Reporter: Chetan Mehrotra
> Assignee: Chetan Mehrotra
> Priority: Minor
> Fix For: 1.3.4
>
> Attachments: OAK-2853.patch
>
>
> Currently OakCodec is used by default if full text indexing is enabled for
> that index. OakCodec disables compression and was done as performance issues
> were observed around 1.0 release (See OAK-1737).
> Post 1.0 we introduced CopyOnRead which should provide better performance
> even with compression enabled. We should revisit the usage of OakCodec by
> default to see if with default code we get comparable performance or not and
> hence get benefit of smaller index size.
> Changing the default would require change in index format version as this
> change would not be compatible to default
> Note that one can still change codec by specifying {{codec}} value for index
> config to the code name like {{Lucene46}}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)