hadoop-pig-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Richard Ding (JIRA)" <j...@apache.org>
Subject [jira] Commented: (PIG-1348) PigStorage making unnecessary byte array copy when storing data
Date Wed, 07 Apr 2010 17:26:34 GMT

    [ https://issues.apache.org/jira/browse/PIG-1348?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12854608#action_12854608
] 

Richard Ding commented on PIG-1348:
-----------------------------------

Thanks Ashutosh. I changed signature of write() to take values of type Tuple instead of type
Object. 

On 1) and 3), Hadoop LineRecordWriter#write() is a synchronized method, and I think that JVM
is optimized for 'instanceof'' construct and also for uncontended synchronization. I prefer
that we have some performance numbers before adding optimizations.

> PigStorage making unnecessary byte array copy when storing data
> ---------------------------------------------------------------
>
>                 Key: PIG-1348
>                 URL: https://issues.apache.org/jira/browse/PIG-1348
>             Project: Pig
>          Issue Type: Bug
>          Components: impl
>    Affects Versions: 0.7.0
>            Reporter: Ashutosh Chauhan
>            Assignee: Richard Ding
>             Fix For: 0.7.0
>
>         Attachments: PIG-1348.patch, PIG-1348_2.patch
>
>
> InternalCachedBag makes estimate of memory available to the VM by using Runtime.getRuntime().maxMemory().
It then uses 10%(by default, though configurable) of this memory and divides this memory into
number of bags. It keeps track of the memory used by bags and then proactively spills if bags
memory usage reach close to these limits. Given all this in theory when presented with data
more then it can handle InternalCachedBag should not run out of memory. But in practice we
find OOM happening. 

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message