[
https://issues.apache.org/jira/browse/FLINK-25646?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17856252#comment-17856252
]
Jufang He edited comment on FLINK-25646 at 6/19/24 11:55 AM:
-------------------------------------------------------------
[~pnowojski] I would like to ask if there is any progress on this issue?
I also saw significant performance degradation when testing buffer debloating
(with Unaligned Checkpoint enabled).
The following is some information about my test jobs. The test kafka source QPS
is high, test jobs are all under backpressure and have lag. With buffer
debloating enabled, the buffer size is smaller, the number of segments used is
increased, and the total amount of in-flight data is significantly decreased,
but the negative effect is also obvious, the throughput of the task is
decreased by more than 30%.
|| ||*Buffer Debloating Disabled*||*Buffer Debloating Enabled*||
|parallelism|1350|1350|
|checkpoint duration(avg)|2m 43s|2m 31s|
|channel state size(avg)|65.9 GB|5.43 GB|
|memory segment size|32K(default value)|256~5k (calculated by buffer
debloating, most of them are 256)|
|total memory segments per TM|115k|115k|
|available memory segments per TM|63.2k|19.5k|
|floating buffers per gate|8(default value)|2000|
|*throughput(avg)*|374k|249k|
was (Author: JIRAUSER302059):
[~pnowojski] I would like to ask if there is any progress on this issue?
I also saw significant performance degradation when testing buffer debloating
(with Unaligned Checkpoint enabled).
The following is some information about my test jobs. The test kafka source QPS
is high, test jobs are all under backpressure and have lag. With buffer
debloating enabled, the buffer size is smaller, the number of segments used is
increased, and the total amount of in-flight data is significantly decreased,
but the negative effect is also obvious, the throughput of the task is
decreased by more than 30%.
|| ||*Buffer Debloating Disabled*||*Buffer Debloating Enabled*||
|parallelism|1350|1350|
|checkpoint duration(avg)|2m 43s|2m 31s|
|channel state size(avg)|65.9 GB|5.43 GB|
|memory segment size|32K(default value)|256~5k (calculated by buffer
debloating, most of them are 256)|
|total memory segments per TM|115k|115k|
|available memory segments per TM|63.2k|19.5k|
|floating buffers per gate|8(default value)|2000|
|*throughput(avg)*|374k|249k|
> Document buffer debloating issues with high parallelism
> -------------------------------------------------------
>
> Key: FLINK-25646
> URL: https://issues.apache.org/jira/browse/FLINK-25646
> Project: Flink
> Issue Type: Improvement
> Components: Documentation, Runtime / Network
> Affects Versions: 1.14.0
> Reporter: Anton Kalashnikov
> Assignee: Anton Kalashnikov
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.15.0
>
>
> According to last benchmarks, there are some problems with buffer debloat
> when job has high parallelism. The high parallelism means the different value
> from job to job but in general it is more than 200. So it makes sense to
> document that problem and propose the solution - increasing the number of
> buffers.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
