You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Piotr Nowojski (Jira)" <ji...@apache.org> on 2020/03/18 06:56:00 UTC

[jira] [Closed] (FLINK-9142) Lower the minimum number of buffers for incoming channels to 1

     [ https://issues.apache.org/jira/browse/FLINK-9142?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Piotr Nowojski closed FLINK-9142.
---------------------------------
    Fix Version/s:     (was: 1.7.3)
       Resolution: Duplicate

I'm closing this ticket, as it was duplicated and superseded by FLINK-16012 and this looks inactive.

> Lower the minimum number of buffers for incoming channels to 1
> --------------------------------------------------------------
>
>                 Key: FLINK-9142
>                 URL: https://issues.apache.org/jira/browse/FLINK-9142
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Runtime / Network
>    Affects Versions: 1.5.0, 1.6.0
>            Reporter: Nico Kruber
>            Assignee: boshu Zheng
>            Priority: Major
>              Labels: pull-request-available
>          Time Spent: 20m
>  Remaining Estimate: 0h
>
> Even if we make the floating buffers optional, we still require {{taskmanager.network.memory.buffers-per-channel}} number of (exclusive) buffers per incoming channel with credit-based flow control while without, the minimum was 1 and only the maximum number of buffers was influenced by this parameter.
> {{taskmanager.network.memory.buffers-per-channel}} is set to {{2}} by default with the argumentation that this way we will have one buffer available for netty to process while a worker thread is processing/deserializing the other buffer. While this seems reasonable, it does increase our minimum requirements. Instead, we could probably live with {{1}} exclusive buffer and up to {{gate.getNumberOfInputChannels() * (networkBuffersPerChannel - 1) + extraNetworkBuffersPerGate}} floating buffers. That way we will have the same memory footprint as before with only slightly changed behaviour.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)