[
https://issues.apache.org/jira/browse/HADOOP-1867?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12526218
]
Doug Cutting commented on HADOOP-1867:
--------------------------------------
> Isn't memory allowed for JVM already available?
Good point. So perhaps we should base things on Runtime#maxMemory()? We could
set the JVM heap size based on /proc/meminfo, then base things on that.
Parameters whose values might be automatically determined include:
- fs.inmemory.size.mb
- io.sort.factor
- io.sort.mb
- io.file.buffer.size
What about mapred.child.java.opts?
> use single parameter to specify a node's available ram
> ------------------------------------------------------
>
> Key: HADOOP-1867
> URL: https://issues.apache.org/jira/browse/HADOOP-1867
> Project: Hadoop
> Issue Type: Improvement
> Reporter: Doug Cutting
>
> To simplify configuration, we should use a single parameter to indicate a
> node's available RAM. Sites should not need to adjust more than this single
> parameter to configure nodes available memory. In task JVMs, some
> significant percentage of the memory should be reserved for application code,
> with the remainder divided among various system buffers.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
