[
https://issues.apache.org/jira/browse/ASTERIXDB-2133?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chen Luo resolved ASTERIXDB-2133.
---------------------------------
Resolution: Fixed
> Unnecessary BinarySearch in GroupFrameAccessor
> ----------------------------------------------
>
> Key: ASTERIXDB-2133
> URL: https://issues.apache.org/jira/browse/ASTERIXDB-2133
> Project: Apache AsterixDB
> Issue Type: Bug
> Components: HYR - Hyracks
> Reporter: Chen Luo
> Assignee: Chen Luo
> Priority: Major
>
> During the merge step of merge sort, if there is enough memory but only a few
> of runs to be merged, we would load multiple frames per run into the
> GroupFrameAccessor. Every time when we access a tuple, GroupFrameAccessor
> performs binary search over the inner frames to translate logical tuple index
> into the physical one (inner frame Id + index).
> However, this is highly inefficient, and partially results in the fact that
> more memory budget of the sort operation would result in slower performance.
> Since GroupFrameAccessor is only used by merge sort, it is expected that
> tuples are accessed sequentially, instead of randomly. Specially
> optimizations can be adopted based on this sequentially access pattern.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
