You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@phoenix.apache.org by "Lars Hofhansl (JIRA)" <ji...@apache.org> on 2014/12/05 06:19:12 UTC
[jira] [Commented] (PHOENIX-1509) Investigate changing default
encoding to PREFIX_encoding instead of FAST_DIFF
[ https://issues.apache.org/jira/browse/PHOENIX-1509?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14235104#comment-14235104 ]
Lars Hofhansl commented on PHOENIX-1509:
----------------------------------------
The main difference between the two is that PREFIX only encoded the prefix of the rowkey+timestamp, whereas FAST_DIFF also finds common bits in the value part of the KV.
For Phoenix we probably mostly want to equalize the overhead from repeating the rowkey and timestamp for every materialized column in a row. So from that angle it would be nice, since it is fast.
If there are many repeated values in *consecutive* KVs then FAST_DIFF is better. Would be awesome to test this out in terms of compression and performance.
> Investigate changing default encoding to PREFIX_encoding instead of FAST_DIFF
> -----------------------------------------------------------------------------
>
> Key: PHOENIX-1509
> URL: https://issues.apache.org/jira/browse/PHOENIX-1509
> Project: Phoenix
> Issue Type: Test
> Reporter: James Taylor
>
> [~lhofhansl] mentioned to me a performance evaluation that found PREFIX_ENCODING to be much less CPU intensive than FAST_DIFF (which is the default for Phoenix tables). We should perf test this ourselves and change the default if the results look good.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)