[
https://issues.apache.org/jira/browse/CASSANDRA-20935?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18121044#comment-18121044
]
Stefan Miklosovic commented on CASSANDRA-20935:
-----------------------------------------------
I think this would be interesting to add as another way of gathering data to
train on during auto-training. The autotraining PR as presented in
CASSANDRA-20937 works only for TWCS tables, looking into last n windows and
training on that. If we were gathering samples continuously as they come to the
database, then we might expand this to other compaction strategies (or it might
be an additional way to gather samples from).
> Add sampling strategies and reservoir for ZSTD dictionary training
> ------------------------------------------------------------------
>
> Key: CASSANDRA-20935
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20935
> Project: Apache Cassandra
> Issue Type: New Feature
> Reporter: Yifan Cai
> Assignee: Warre Pessers
> Priority: Normal
>
> Introduce support for sampling strategies in ZSTD dictionary auto-training
> and persist sampled data in a reservoir for reuse.
> - Support: random sampling and exponential decay reservoir sampling.
> - Maintain a reservoir of sampled data for consistent and representative
> training.
> - Ensure integration between sampling and reservoir storage is seamless.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
