[
https://issues.apache.org/jira/browse/FLINK-4545?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15924535#comment-15924535
]
Nico Kruber commented on FLINK-4545:
------------------------------------
We did some thinking and would probably add the following three new
configuration parameters (with the given defaults) to finally replace the
{{taskmanager.network.numberOfBuffers}} parameter:
- {{taskmanager.network.memory.fraction}} (default: 0.1): fraction of JVM
memory to use for network buffers (by reducing {{taskmanager.memory.fraction}}
from 0.7 to 0.6)
- {{taskmanager.network.memory.min}} (default: 64MB): minimum memory size for
network buffers
- {{taskmanager.network.memory.max}} (default: 1GB): maximum memory size for
network buffers
A fixed size may be achieved by setting the latter two to the same value,
{{taskmanager.network.numberOfBuffers}} will be marked deprecated and used only
if the other three are not given, e.g. due to old config files being used.
> Flink automatically manages TM network buffer
> ---------------------------------------------
>
> Key: FLINK-4545
> URL: https://issues.apache.org/jira/browse/FLINK-4545
> Project: Flink
> Issue Type: Wish
> Components: Network
> Reporter: Zhenzhong Xu
>
> Currently, the number of network buffer per task manager is preconfigured and
> the memory is pre-allocated through taskmanager.network.numberOfBuffers
> config. In a Job DAG with shuffle phase, this number can go up very high
> depends on the TM cluster size. The formula for calculating the buffer count
> is documented here
> (https://ci.apache.org/projects/flink/flink-docs-master/setup/config.html#configuring-the-network-buffers).
>
> #slots-per-TM^2 * #TMs * 4
> In a standalone deployment, we may need to control the task manager cluster
> size dynamically and then leverage the up-coming Flink feature to support
> scaling job parallelism/rescaling at runtime.
> If the buffer count config is static at runtime and cannot be changed without
> restarting task manager process, this may add latency and complexity for
> scaling process. I am wondering if there is already any discussion around
> whether the network buffer should be automatically managed by Flink or at
> least expose some API to allow it to be reconfigured. Let me know if there is
> any existing JIRA that I should follow.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
