hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-5000) Speed up simultaneous reads of a block when block caching is turned off
Date Mon, 12 Dec 2011 18:35:30 GMT

    [ https://issues.apache.org/jira/browse/HBASE-5000?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13167655#comment-13167655
] 

stack commented on HBASE-5000:
------------------------------

I added a bit of rough doc suggesting users not disable blockcache completely.

{code}
         <section xml:id="disabling.blockcache"><title>Disabling Blockcache</title>
           <para>Do not turn off block cache (You'd do it by setting <varname>hbase.block.cache.size</varname>
to zero).
           Currently we do not do well if you do this because the regionserver will spend
all its time loading hfile
           indices over and over again.  If your working set it such that block cache does
you no good, at least
           size the block cache such that hfile indices will stay up in the cache (you can
get a rough idea
           on the size you need by surveying regionserver UIs; you'll see index block size
accounted near the
           top of the webpage).</para>
         </section>
{code}
                
> Speed up simultaneous reads of a block when block caching is turned off
> -----------------------------------------------------------------------
>
>                 Key: HBASE-5000
>                 URL: https://issues.apache.org/jira/browse/HBASE-5000
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Mikhail Bautin
>            Assignee: Mikhail Bautin
>            Priority: Minor
>
> With block caching, when one client starts reading a block and another one comes around
asking for the same block, the second client waits for the first one to finish reading and
returns the block from cache. This is achieved by locking on the block offset using IdLock,
a "sparse lock" primitive allowing to lock on arbitrary long numbers. However, in case there
is no block caching, there is no reason to wait for other clients that are reading the same
block. One challenge optimizing this that we don't necessary have accurate information about
whether other HFile API clients interested in the block would cache it.
> Setting priority as minor, as it is very unusual to turn off block caching.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message