[
https://issues.apache.org/jira/browse/CASSANDRA-9786?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Benedict updated CASSANDRA-9786:
--------------------------------
Labels: performance (was: )
> Our index file seeks can be directed to the page most likely to contain the
> record, and read exactly one page
> -------------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-9786
> URL: https://issues.apache.org/jira/browse/CASSANDRA-9786
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Benedict
> Labels: performance
> Fix For: 3.x
>
>
> (Perhaps two, if it is likely to cross a page boundary).
> This only works if the partitioner is hash based, but if so we can expect the
> keys to be uniformly distributed, so we can easily calculate the likelihood
> of finding it on a given page. With CASSANDRA-8931, we can also expect the
> data to be of a uniform size.
> This will require a little bit of index file restructuring, so that each page
> can be read as a single unit. A follow-up ticket may then permit in-memory
> binary search within the page for the desired record (with variable key sizes
> this may be difficult, so perhaps a follow-up).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
