[
https://issues.apache.org/jira/browse/CASSANDRA-8988?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14368959#comment-14368959
]
Branimir Lambov commented on CASSANDRA-8988:
--------------------------------------------
Nicely done.
[AsymmetricOrdering
76|https://github.com/apache/cassandra/compare/trunk...belliottsmith:8988#diff-e8d8865e0c7d14541ae677a1c98bfca4R76]:
I think this will be clearer if it follows the structure of the one above:
{{if (c < 0) i = m; else j = m;}} to make it obvious that the only difference
between the two is the strictness of the comparison.
[IntervalTree
235|https://github.com/apache/cassandra/compare/trunk...belliottsmith:8988#diff-e675fa5966322284415eff48ec0b36ffR235]:
You should use CEIL (which is the same as LOWER + 1) here. It may be worth
documenting in the Op declaration that CEIL == LOWER + 1, HIGHER == FLOOR + 1
(this is true because j is always equal to i+1 when exiting the loop in find2).
[IntervalTree
236|https://github.com/apache/cassandra/compare/trunk...belliottsmith:8988#diff-e675fa5966322284415eff48ec0b36ffR236],
[248|https://github.com/apache/cassandra/compare/trunk...belliottsmith:8988#diff-e675fa5966322284415eff48ec0b36ffR248]:
I'm worried that this may be increasing the complexity over the original code
for bigger trees as it does one more level of intersectsX search. It's your
call, but my preference is to do the outside span rejection before the search
as otherwise there is a possibility of performance regression from this,
however small.
> Optimise IntervalTree
> ---------------------
>
> Key: CASSANDRA-8988
> URL: https://issues.apache.org/jira/browse/CASSANDRA-8988
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Benedict
> Assignee: Benedict
> Priority: Trivial
> Fix For: 2.1.4
>
> Attachments: 8988.txt
>
>
> We perform a lot of unnecessary comparisons in
> IntervalTree.IntervalNode.searchInternal.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
