You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@giraph.apache.org by Jonathan Bishop <jb...@gmail.com> on 2012/11/27 19:07:17 UTC
Minimum superstep time
Hi,
I am involved in a project which requires thousands of supersteps. In each
superstep there are about 1 thousand vertices, each sending a message. The
message size is not large, about 40 bytes or so.
I am seeing abouth 3000-5000ms per superstep. I am curious what others with
similar problem sizes are seeing.
Thanks,
Jon
Re: Minimum superstep time
Posted by Sebastian Schelter <ss...@apache.org>.
On 27.11.2012 22:53, Jonathan Bishop wrote:
> Sebastian,
>
> I am evaluating a directed acyclic graph (DAG). So the vertices in one
> superstep are not dependent on each other, only upon their predecessors.
That's what I meant, the active vertices in step (t+1) depend on those
of step (t), so you can't run these simultaneously.
Can you say what problem you are trying to solve? And how big your graph is?
/s
>
> What do you mean "execute the messaging simultaneously"? I am simply using
> BasicVertex.sendMsg() during BasicVertex.compute(). Is there another way to
> do this?
>
> Jon
>
>
>
>
> On Tue, Nov 27, 2012 at 10:19 AM, Sebastian Schelter <ss...@apache.org> wrote:
>
>> Are the vertices which are active in supersteps dependent on each other?
>> If this is not the case, you could try to execute the messaging
>> simultaneously.
>>
>> Could you give a little more details about the problem, which you are
>> trying to solve?
>>
>> /s
>>
>>
>> On 27.11.2012 19:07, Jonathan Bishop wrote:
>>> Hi,
>>>
>>> I am involved in a project which requires thousands of supersteps. In
>> each
>>> superstep there are about 1 thousand vertices, each sending a message.
>> The
>>> message size is not large, about 40 bytes or so.
>>>
>>> I am seeing abouth 3000-5000ms per superstep. I am curious what others
>> with
>>> similar problem sizes are seeing.
>>>
>>> Thanks,
>>>
>>> Jon
>>>
>>
>>
>
Re: Minimum superstep time
Posted by Jonathan Bishop <jb...@gmail.com>.
Sebastian,
I am evaluating a directed acyclic graph (DAG). So the vertices in one
superstep are not dependent on each other, only upon their predecessors.
What do you mean "execute the messaging simultaneously"? I am simply using
BasicVertex.sendMsg() during BasicVertex.compute(). Is there another way to
do this?
Jon
On Tue, Nov 27, 2012 at 10:19 AM, Sebastian Schelter <ss...@apache.org> wrote:
> Are the vertices which are active in supersteps dependent on each other?
> If this is not the case, you could try to execute the messaging
> simultaneously.
>
> Could you give a little more details about the problem, which you are
> trying to solve?
>
> /s
>
>
> On 27.11.2012 19:07, Jonathan Bishop wrote:
> > Hi,
> >
> > I am involved in a project which requires thousands of supersteps. In
> each
> > superstep there are about 1 thousand vertices, each sending a message.
> The
> > message size is not large, about 40 bytes or so.
> >
> > I am seeing abouth 3000-5000ms per superstep. I am curious what others
> with
> > similar problem sizes are seeing.
> >
> > Thanks,
> >
> > Jon
> >
>
>
Re: Minimum superstep time
Posted by Sebastian Schelter <ss...@apache.org>.
Are the vertices which are active in supersteps dependent on each other?
If this is not the case, you could try to execute the messaging
simultaneously.
Could you give a little more details about the problem, which you are
trying to solve?
/s
On 27.11.2012 19:07, Jonathan Bishop wrote:
> Hi,
>
> I am involved in a project which requires thousands of supersteps. In each
> superstep there are about 1 thousand vertices, each sending a message. The
> message size is not large, about 40 bytes or so.
>
> I am seeing abouth 3000-5000ms per superstep. I am curious what others with
> similar problem sizes are seeing.
>
> Thanks,
>
> Jon
>