cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benedict (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-8896) Investigate upstream changes to compressors to fit contents exactly to one page
Date Wed, 04 Mar 2015 16:26:04 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-8896?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14347082#comment-14347082
] 

Benedict commented on CASSANDRA-8896:
-------------------------------------

It's not something I'm personally likely to tackle in the next month or so, as I have a number
of other things no my plate. But I think it is a worthwhile improvement to sneak into 3.0
if we can, as it's relatively well contained but should result in more predictable disk (and
page cache) performance, especially alongside dynamically selecting the compressed page size
based on compression ratio.

> Investigate upstream changes to compressors to fit contents exactly to one page
> -------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-8896
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-8896
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Core
>            Reporter: Benedict
>
> For optimal disk performance, it makes most sense to choose our compression boundaries
based on compressed size, not uncompressed. If our compressors could take a target length,
and return the number of source bytes they managed to fit into that space, this would permit
us to lower the number of disk accesses per read. [~blambov]: you've dived into LZ4. How tricky
do you think this might be?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message