[
https://issues.apache.org/jira/browse/CASSANDRA-3378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13564898#comment-13564898
]
Jason Brown commented on CASSANDRA-3378:
----------------------------------------
Looks like snappy does it's own internal buffering (for compressing streams).
SnappyOutputStream has an 'uncompressed' buffer of 32k, which flushes to it's
output stream when full or flush() is called. SnappyInputStream creates a
buffer based upon the size of the chunk size of the current block. Thus, we
shouldn't need to worry about having Buffered*putStream for compressed streams.
We could play with the buffer size there (4k vs. 8k), but I think the biggest
win is just having _any_ buffer before the socket output stream (which doesn't
buffer at all). I'm in favor of leaving the default 4096 buffer size for
Buffered*putStream on uncompressed streams until there is sufficient
evidence/testing to switch.
As for the socket's buffer config, we can use Michal's new yaml settings for
the in/out sockets. However, I think storage_send_buff_size_in_bytes should be
commented out by default in the yaml.
> Allow configuration of storage protocol socket buffer
> -----------------------------------------------------
>
> Key: CASSANDRA-3378
> URL: https://issues.apache.org/jira/browse/CASSANDRA-3378
> Project: Cassandra
> Issue Type: New Feature
> Components: Core
> Reporter: Brandon Williams
> Assignee: Michał Michalski
> Priority: Minor
> Labels: lhf
> Attachments: cassandra-3378-v1.patch, cassandra-3378-v2.patch
>
>
> Similar to rpc_[send,recv]_buff_size_in_bytes, we should expose this for high
> latency connections.
--
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
