[
https://issues.apache.org/jira/browse/IGNITE-26007?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18011147#comment-18011147
]
Roman Puchkovskiy commented on IGNITE-26007:
--------------------------------------------
Now, the corresponding pool gets CPUs * 2 threads which is usually already too
much. If we just replicate it, it will be 4 * CPUs threads in total.
I believe we should only do this after we solve the 'reasonable network workers
pool size' problem (IGNITE-24415 might be related). But we'll probably need to
choose different ways to detect the default pool size: for server2server, the
anticipated cluster size might make sense to start with, while for client
handler, it should be the anticipated number of clients per server node.
Guys working on the Client Handler should probably be consulted first.
> Divide IO threads between clients and servers
> ---------------------------------------------
>
> Key: IGNITE-26007
> URL: https://issues.apache.org/jira/browse/IGNITE-26007
> Project: Ignite
> Issue Type: Improvement
> Reporter: Vladislav Pyatkov
> Priority: Major
> Labels: ignite-3
>
> h3. Motivation
> Clients and servers have a different load profile. They can interfere with
> each other.
> h3. Definition of done
> The client has a dedicated range of IO threads and cannot coalesce with a
> server one.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
