[
https://issues.apache.org/jira/browse/GIRAPH-12?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hyunsik Choi updated GIRAPH-12:
-------------------------------
Attachment: GIRAPH-12_3.patch
Avery,
Thank you for your comments. Yes, I agreed that the aggregated wasted memory
should be considered =)
In the latest comment, I investigated the real occupation of thread stack by
using Sleep class (https://gist.github.com/1249761). When I created 2000 Sleep
threads with the default stack size option '-Xss4096k', the memory usage of
both the process and all threads is only 46 mega bytes.
So, I would like to say that the individual thread consumes much less stack
size than default thread stack size. The default thread stack size affects the
virtual memory area size. It is not resident memory size. The actual stack size
per thread seems to be only affected by local variables and function
invocations.
As a result, I guess that the memory problem is usually caused by outgoing
messages kept in memory =)
Anyway, I attach the patch. The main difference from the previous patch is that
the default number of thread pool is set to the number of workers - 1 if
unset. Besides, I added more comments.
The unit tests are passed against the real hadoop cluster.
> Investigate communication improvements
> --------------------------------------
>
> Key: GIRAPH-12
> URL: https://issues.apache.org/jira/browse/GIRAPH-12
> Project: Giraph
> Issue Type: Improvement
> Components: bsp
> Reporter: Avery Ching
> Assignee: Hyunsik Choi
> Priority: Minor
> Attachments: GIRAPH-12_1.patch, GIRAPH-12_2.patch, GIRAPH-12_3.patch
>
>
> Currently every worker will start up a thread to communicate with every other
> workers. Hadoop RPC is used for communication. For instance if there are
> 400 workers, each worker will create 400 threads. This ends up using a lot
> of memory, even with the option
> -Dmapred.child.java.opts="-Xss64k".
> It would be good to investigate using frameworks like Netty or custom roll
> our own to improve this situation. By moving away from Hadoop RPC, we would
> also make compatibility of different Hadoop versions easier.
--
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
