[
https://issues.apache.org/jira/browse/TIKA-2645?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16482862#comment-16482862
]
Tim Allison commented on TIKA-2645:
-----------------------------------
XML root detection before (time in millis)...with the pool size equal to the
thread count in both tests.
20 Threads:
||Run||No Pool||With Pool||
|1|5897|3165|
|2|5548|3130|
|3|5506|3105|
Single thread:
||Run||No Pool||With Pool||
|1|1790|912|
|2|1737|863|
|3|1702|861|
> Reuse SAXParsers where possible
> -------------------------------
>
> Key: TIKA-2645
> URL: https://issues.apache.org/jira/browse/TIKA-2645
> Project: Tika
> Issue Type: Task
> Reporter: Tim Allison
> Priority: Major
>
> [~wastl-nagel] pointed out on the [user
> list|https://lists.apache.org/thread.html/54913f29cd83bba175f77a2d6a4902bb3a5cba2fa495bbfd6012024a@%3Cuser.tika.apache.org%3E]
> that there can be thread contention when creating SAXParsers. [~jukkaz]
> recommended pooling SAXParsers.
> I noticed a modest improvement in speed when parsing docx files with our SAX
> based parser, and I suspect that [~wastl-nagel] will find an even greater
> improvement during detection because we're currently creating a new SAXParser
> on every call to detect an XML root.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
