[ 
https://issues.apache.org/jira/browse/HBASE-15316?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15284911#comment-15284911
 ] 

Elliott Clark commented on HBASE-15316:
---------------------------------------

bq.When write block, we cache ROOT_INDEX by default
I mean that we should have the ability to cache ROOT_INDEX on writing a HFile. 
Right now we have the ability to cache all index blocks on write. However as 
data gets big that can be a lot of data and might include very stale indices. 
Root index however is read many many times and should be pretty small.

> Add the ability to cache ROOT_INDEX on reads only
> -------------------------------------------------
>
>                 Key: HBASE-15316
>                 URL: https://issues.apache.org/jira/browse/HBASE-15316
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Elliott Clark
>
> For larger data sets it's sometimes not needed to fit the entire index in 
> memory ( there can be cold spots in the data, too much data in index, etc). 
> However it's almost always needed to cache the root index. It's always the 
> hottest blocks of the entire region.
> We should allow the root index block to be cached in memory on write. We 
> should also do that by default.



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

Reply via email to