You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hama.apache.org by wangzhigang1210 <wa...@163.com> on 2013/09/04 02:28:41 UTC

for graph processing, hama is disk-resident or memory-resident?

Hi,
Recently, I read source code of Hama on Graph API. And I have a question about the disk-resident and memory-resident issue.
I find that there are two classes for maintaining vertices: ListVerticesInfo.java and DiskVerticesInfo.java. The latter supports disk operations. However, in the class of GraphJobRunner.java, the class is set as ListVerticesInfo in the function setupFields. Thus, it seems that the version of disk is unusable for Graph API.
Similarly, in the function of GraphJob.submit(), the version of message queue is also set as SortedMessageQueue instead of DiskQueue.
So, for graph processing, I want to know whether Hama is a disk-resident system or not.
Thanks,
Zhigang Wang

Re: for graph processing, hama is disk-resident or memory-resident?

Posted by "Edward J. Yoon" <ed...@apache.org>.
Hi,

You cannot use the disk-based vertices storage and queue at the moment
but it will be supported in the future. We're under heavy development.


On Wed, Sep 4, 2013 at 9:28 AM, wangzhigang1210 <wa...@163.com> wrote:
> Hi,
> Recently, I read source code of Hama on Graph API. And I have a question about the disk-resident and memory-resident issue.
> I find that there are two classes for maintaining vertices: ListVerticesInfo.java and DiskVerticesInfo.java. The latter supports disk operations. However, in the class of GraphJobRunner.java, the class is set as ListVerticesInfo in the function setupFields. Thus, it seems that the version of disk is unusable for Graph API.
> Similarly, in the function of GraphJob.submit(), the version of message queue is also set as SortedMessageQueue instead of DiskQueue.
> So, for graph processing, I want to know whether Hama is a disk-resident system or not.
> Thanks,
> Zhigang Wang



-- 
Best Regards, Edward J. Yoon
@eddieyoon