[
https://issues.apache.org/jira/browse/CASSANDRA-950?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12853527#action_12853527
]
Jonathan Ellis commented on CASSANDRA-950:
------------------------------------------
I'd be more comfortable if we only cache the length on init for read-only BRAF
modes (thus not having to add any code to the inner loop of our write path).
Incidentally, I found this while googling, indicating that length() makes not
one but three system calls per invocation. So this seems like a worthwhile
optimization.
http://blogs.oracle.com/charlesLamb/2008/09/javaiorandomaccessfilelength.html
> Cache BufferedRandomAccessFile.length
> -------------------------------------
>
> Key: CASSANDRA-950
> URL: https://issues.apache.org/jira/browse/CASSANDRA-950
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Johan Oskarsson
> Attachments: CASSANDRA-950.patch
>
>
> Profiling shows that BufferedRandomAccessFile.length is eating up an
> unexpected amount of cpu. It believe it is the call through the "native
> barrier" that is causing it. We often call the length method via is end of
> file when reading data. We can add caching of the file length to avoid this
> call most times.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
