hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ramkrishna.s.vasudevan (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-17353) Cache famLen in OffheapKeyValue
Date Wed, 21 Dec 2016 05:10:58 GMT

    [ https://issues.apache.org/jira/browse/HBASE-17353?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15766148#comment-15766148
] 

ramkrishna.s.vasudevan commented on HBASE-17353:
------------------------------------------------

We will have one more byte heap overhead for an offheap cell. We have now 47 bytes FIXED overhead
for an offheap kv. Now if we cache famLen it will be 48 bytes. Not a big change but still
when we have lot of small cells then we have a price for heap overhead though the cell data
is offheap.

> Cache famLen in OffheapKeyValue
> -------------------------------
>
>                 Key: HBASE-17353
>                 URL: https://issues.apache.org/jira/browse/HBASE-17353
>             Project: HBase
>          Issue Type: Improvement
>    Affects Versions: 2.0.0
>            Reporter: ramkrishna.s.vasudevan
>            Assignee: ramkrishna.s.vasudevan
>            Priority: Minor
>             Fix For: 2.0.0
>
>         Attachments: OffheapKV famLen cost.png
>
>
> WE need to discuss here. Already we had a TODO here. But this comes again after offheap
memstore is committed in trunk. 
> Attaching a screenshot to show the impact. True that these changes won't have a direct
impact on the final perf but at micro level they would have. 
> In case of KeyValue it is just retrieving a byte from the byte[] (o(1) access).
> But here we need to access the memory to retrive that one byte though it is Unsafe based.



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

Mime
View raw message