[
https://issues.apache.org/jira/browse/CASSANDRA-10355?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14877155#comment-14877155
]
Benedict commented on CASSANDRA-10355:
--------------------------------------
I had an (incomplete) hack at this on the plane, and while [the benefit is
real|http://cstar.datastax.com/graph?stats=f2d98688-5e7a-11e5-bf84-42010af0688f],
it is not sufficient to make a difference for 3.0.
> Flatten DataInputPlus type hierarchy
> ------------------------------------
>
> Key: CASSANDRA-10355
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10355
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Benedict
> Assignee: Benedict
> Fix For: 3.x
>
>
> Almost every instance extends {{RebufferingInputStream}} now, however there
> are two last implementations hanging around that we can probably do without.
> If we do so, the majority of method calls to any {{DataInputPlus}} (i.e., the
> majority of method calls full stop) can be inlined by hotspot. Thus giving us
> behaviour pretty similar to passing around raw {{ByteBuffer}} objects.
> It's worth a shot, anyway.
> If it works remotely well, we should consider splitting readUnsignedVInt into
> its first byte read (which we would aim to inline) and its proceeding larger
> read (which we would perhaps ask not to be).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
