[
https://issues.apache.org/jira/browse/IOTDB-981?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17225604#comment-17225604
]
Xiangdong Huang commented on IOTDB-981:
---------------------------------------
The idea works, I think.
But maybe we will go to the opposite extreme, i.e., all chunk metadata caches
are unseq metaedata..
If so, will it cause some problems?
> Give higher priority to unseq data in chunk cache
> -------------------------------------------------
>
> Key: IOTDB-981
> URL: https://issues.apache.org/jira/browse/IOTDB-981
> Project: Apache IoTDB
> Issue Type: Improvement
> Components: Core/Engine
> Reporter: Tian Jiang
> Priority: Minor
>
> Cache unseq data chunk may provide large benefits for queries, consider the
> following situation:
> We have 3 seq chunks: s0, s1, and s2; and 1 unseq chunk: u0, which overlaps
> with all 3 seq chunks. Assuming the 3 chunks are stored more closely (having
> lowing seek costs when accessed sequentially).
> If we can only cache one chunk in chunk cache, and if we choose to cache any
> of the seq chunks, we then must seek between seq chunks and unseq chunks. If
> we cache the unseq chunk, then only seeks between seq chunks are required,
> which is clearly more efficient.
> The simple example shows that giving a higher priority to unseq chunks is
> beneficial, but it remains to be seen whether this conclusion can be
> generalized.
> Any further discussion is welcomed here.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
