You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@hama.apache.org by Andronidis Anastasios <an...@hotmail.com> on 2014/09/24 11:30:34 UTC

Future plans

I am thinking on working on a VertexInfo instance, that will not serialize-deserialize the vertices all the time for better performance if there is enough memory.

Also, I want to rework the aggregators to be able to pass arbitrary messages and not stay bound to Vertex-Message type.

Anastasis

Re: Future plans

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

On Wed, Sep 24, 2014 at 6:30 PM, Andronidis Anastasios
<an...@hotmail.com> wrote:
> I am thinking on working on a VertexInfo instance, that will not serialize-deserialize the vertices all the time for better performance if there is enough memory.

+1

> Also, I want to rework the aggregators to be able to pass arbitrary messages and not stay bound to Vertex-Message type.

I personally think that Aggregators are somewhat complex. If we use
the barrier sync for Aggregators, two times number of original
supersteps is needed and vertex messages need to be sent to S + 2. To
avoid meaningless add and poll operations, we currently uses the
persistent queue. In fact, I have no idea at this moment.

> Anastasis

-- 
Best Regards, Edward J. Yoon
CEO at DataSayer Co., Ltd.