[
https://issues.apache.org/jira/browse/MINIFICPP-1538?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17324910#comment-17324910
]
Adam Debreceni commented on MINIFICPP-1538:
-------------------------------------------
Current progress
[here.|https://github.com/adamdebreceni/nifi-minifi-cpp/tree/MINIFICPP-1538]
Based on measurements on Windows machine, during flow operations there is no
discernible benefit using per-process-group column families, when the number of
parallel process groups is 2, both of them using a [ConsumeWindowsEventLog ->
LogAttribute] setup
!Screenshot 2021-04-14 at 12.50.05.png|width=797,height=535!
!Screenshot 2021-04-14 at 12.50.14.png|width=807,height=538!
On synthetic benchmarks continuously reading/writing in 2 parallel "flows" each
one having 4 readers and 4 writers the benefit is measurable:
windows:
* with column families: ~470k
* without: ~300k
mac:
* with column families: ~710k
* without: ~620k
Based on these measurements the complexity/amount of code required for this
feature outweighs the provided benefits.
> Investigate per process group column families
> ---------------------------------------------
>
> Key: MINIFICPP-1538
> URL: https://issues.apache.org/jira/browse/MINIFICPP-1538
> Project: Apache NiFi MiNiFi C++
> Issue Type: New Feature
> Reporter: Adam Debreceni
> Assignee: Adam Debreceni
> Priority: Major
> Attachments: Screenshot 2021-04-14 at 12.50.05.png, Screenshot
> 2021-04-14 at 12.50.14.png
>
>
> Currently all operations in FlowFileRepository go through a single column
> family (the default). We should investigate the performance benefit, if any,
> of using different column families for different process groups (special care
> must be taken for RPGs, the processors of which logically belong to their
> parent groups).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
