[
https://issues.apache.org/jira/browse/YUNIKORN-3323?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18101647#comment-18101647
]
Wilfred Spiegelenburg commented on YUNIKORN-3323:
-------------------------------------------------
Tagging for release note: the performance gain could be large enough that we
want to track it as a major new item
Comparing the output of {{BenchmarkScheduling()}} for:
* old (google based btree)
* new tidwall BtreeG with Scan
* new tidwall BtreeG with Walk
Walk might be "faster" in the benchmark but does it show for us, and if it
shows is the code complexity needed to use it a good trade-off for the speed
gain.
Same output can be used for comparison in the release notes.
> Replace google btree for node structures
> ----------------------------------------
>
> Key: YUNIKORN-3323
> URL: https://issues.apache.org/jira/browse/YUNIKORN-3323
> Project: Apache YuniKorn
> Issue Type: Bug
> Components: core - scheduler
> Reporter: Wilfred Spiegelenburg
> Assignee: Aditya Maheshwari
> Priority: Critical
> Labels: pull-request-available, release-notes
>
> The core uses a btree to keep a sorted structure for the nodes.
> The underlying implementation for the btree is the
> [google/btree|https://github.com/google/btree]. The repository has been
> archived and is no longer supported.
> Moving to a different implementation for a btree that is still supported is
> the correct thing.
> The simple replacement seems to be the
> [tidwall/btree|https://github.com/tidwall/btree]. Moving to the generics
> version inside that the implementation would also reduce memory overhead and
> improve performance.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
