[
https://issues.apache.org/jira/browse/CASSANDRA-5043?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13529247#comment-13529247
]
Pavel Yaskevich commented on CASSANDRA-5043:
--------------------------------------------
bq. ABSC might be faster for 10 columns but there are definitely going to be
queries for which log(N) is going to beat N. It would be nice to quantify this
somehow in terms of what we know (number of sstables and number of columns
being requested for sure; possibly row size and compaction strategy)
name slices are not that big e.g. 15-20 as you mentioned, row size is around
~700KB e.g. 50 columns and compaction strategy used is
SizeTiredCompactionStrategy, switching to ABSC allowed to cut read latencies
almost in half.
> Small CollationController refactoring and CFS.getRawCachedRow(DecoratedKey)
> fix.
> --------------------------------------------------------------------------------
>
> Key: CASSANDRA-5043
> URL: https://issues.apache.org/jira/browse/CASSANDRA-5043
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Pavel Yaskevich
> Assignee: Pavel Yaskevich
> Priority: Minor
> Fix For: 1.2.0
>
> Attachments: 5043-v3.txt,
> ABSC-with-addAll-for-time-ordered-data.patch, CASSANDRA-5043.patch,
> CASSANDRA-5043-v2.patch
>
>
> CollactionController can construct a factory on init based on mutableColumns
> parameter. ArrayBackedSortedColumns proven to be better for counters as well
> as normal columns (row size from 5 to 75 columns). CFS.getRawCachedRow would
> copy key byte array each time it contracts a RowCacheKey so we want to avoid
> that where possible.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
