[
https://issues.apache.org/jira/browse/CASSANDRA-20212?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dmitry Konstantinov updated CASSANDRA-20212:
--------------------------------------------
Change Category: Performance
Complexity: Normal
Component/s: Local/Memtable
> Avoid megamorpthic calls for NativeCell
> ---------------------------------------
>
> Key: CASSANDRA-20212
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20212
> Project: Apache Cassandra
> Issue Type: Improvement
> Components: Local/Memtable
> Reporter: Dmitry Konstantinov
> Assignee: Dmitry Konstantinov
> Priority: Normal
> Attachments: image-2025-01-15-21-31-21-426.png
>
>
> If we use offheap_objects allocation mode we have 3 implementations of Cells
> used at the same time:
> * NativeCell
> * BufferCell
> * ArrayCell (CASSANDRA-15393)
> When all of them are invoked from the same callsite JIT compiler cannot
> optimize such calls properly and we have megamorphic virtual calls with
> correspondent overheads and lack of inlining.
> The basic way to solve this issue is to reduce the number of Cell
> implementations are actively used and invoked from the same callsites.
> CASSANDRA-20166 replaces BufferCell with ArrayCell
> Other places to consider:
> * SimpleBuilders logic
> !image-2025-01-15-21-31-21-426.png|width=300!
> * EnsureOnHeap / HeapCloner used to clone data during reads
> for them probably we can also move to ArrayCell usage.
>
> Related links:
> *
> [http://insightfullogic.com/2014/May/12/fast-and-megamorphic-what-influences-method-invoca/]
>
> *
> [https://mechanical-sympathy.blogspot.com/2012/04/invoke-interface-optimisations.html]
>
> * [https://shipilev.net/blog/2015/black-magic-method-dispatch/]
> * [https://shipilev.net/jvm/anatomy-quarks/16-megamorphic-virtual-calls/]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
