[
https://issues.apache.org/jira/browse/HBASE-7630?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13558115#comment-13558115
]
stack commented on HBASE-7630:
------------------------------
Issues with a large number of regions:
+ Startup takes longer because more regions to assign
+ Recovery could take longer because we have to write more files and assign
more if more regions (Caveat some the change to log splitting proposed by
Jeffrey
+ Each region gets its own MemStore per column family instance in a region so
more regions would mean more of these and currently if mslab enabled, its at
least a 2MB allocation each if an entry..... (hopefully too our flushing will
get better soon; i.e. we don't flush all cfs in a region, just the big ones
when under memory pressure)
If more regions, would be cool if some regions could go cold, not occupy
resources if not accessed in last 24 hours w/ some minimal amount of start up
time once an access came in.
What is the objective Sergey? Serving up all of the attached disk content?
> support large number of small regions
> -------------------------------------
>
> Key: HBASE-7630
> URL: https://issues.apache.org/jira/browse/HBASE-7630
> Project: HBase
> Issue Type: Brainstorming
> Reporter: Sergey Shelukhin
>
> Supporting large number of small regions can help improve scalability in
> terms of number of tables/cfs/etc., help speed up recovery, and also lessen
> the impact of compactions (due to there being a large number of small
> compactions).
> This is an issue to attempt brainstorming current limitations.
> We might want to take this into consideration when designing master tasks
> such as in HBASE-5487.
--
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
