[
https://issues.apache.org/jira/browse/HBASE-8423?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13642091#comment-13642091
]
Nicolas Spiegelberg commented on HBASE-8423:
--------------------------------------------
See: http://svn.apache.org/viewvc?view=revision&revision=1475883
Note that I would definitely consider this an advanced feature and exploratory.
I jury-rigged a way to track Major Compacted file metrics versus non-mc files
to get the stats for this theory. I suggest that only people who know how to
look at HBase metrics use this feature. Not a lot of code, but definitely
diligent analysis necessary. Also, note that I use the Dynamic Schema Config
from HBASE-5335 to change this on a per-CF basis.
> Allow Major Compaction to Use Different Compression
> ---------------------------------------------------
>
> Key: HBASE-8423
> URL: https://issues.apache.org/jira/browse/HBASE-8423
> Project: HBase
> Issue Type: Improvement
> Reporter: Nicolas Spiegelberg
> Priority: Minor
> Labels: compaction, regionserver
>
> For the Messages use case, ~90% of the disk util is associated with major
> compacted files but less than 33% of the get IOPS are due to mc files.
> Allowing a different compaction algorithm for major compacted files will
> allow us to get ~30% disk savings. Latency should be higher, but storefile
> querying is sequential, so only 1/3 of the get latency is from disk access to
> these files.
--
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
