You can certainly implement BSP on top of a MapReduce implementation.
But this is going to be very very expensive. Consider that all
communication in MapReduce will go through the phase of storing map
outputs locally (disk) before being send to the reducer. Also, consider
than the entire gra
Avery,
> Communication between mappers is not part of the MapReduce computing
model. Therefore, it doesn't make sense for them to include it as it would
unnecessarily complicate the fault-tolerance recovery.
I agree that it doesn't make sense to complicate things by introducing
communication bet
> There must have been definitely some thought around this.
Yes, there was some thought put into the design before writing - in
Hadoop's case - hundreds of thousands of lines of code, and, for
Giraph, thousands so far. I invite you to read the MapReduce paper
(http://research.google.com/archive/m
On 12/9/11 10:22 PM, Praveen Sripati wrote:
Jack,
> Giraph maps do communicate: via RPC. This is done repeatedly in
every mapper, during the compute phase. This is something that is not
normal to MapReduce, it is special to Giraph.
There must have been definitely some thought around this.
Jack,
> Giraph maps do communicate: via RPC. This is done repeatedly in every
mapper, during the compute phase. This is something that is not normal to
MapReduce, it is special to Giraph.
There must have been definitely some thought around this. But, we can also
have a mapper correspond to just
On Fri, Dec 9, 2011 at 8:16 PM, Praveen Sripati wrote:
> Jake,
>
>
> > Let's not crosspost, please, it make the thread of conversation totally
> opaque as to who is talking about what.
>
> Agree. I got it after the OP.
>
>
> > There is only one set of map tasks for the Giraph job - those
> long-ru
Jake,
> Let's not crosspost, please, it make the thread of conversation totally
opaque as to who is talking about what.
Agree. I got it after the OP.
> There is only one set of map tasks for the Giraph job - those
long-running map tasks run possibly many supersteps.
OK. But, map tasks don't com
[hama-user to bcc:]
Let's not crosspost, please, it make the thread of conversation totally
opaque as to who is talking about what.
On Fri, Dec 9, 2011 at 1:42 AM, Praveen Sripati wrote:
> Thanks to Thomas and Avery for the response.
>
> > For Giraph you are quite correct, all the stuff is submi
Thanks to Thomas and Avery for the response.
> For Giraph you are quite correct, all the stuff is submitted as a MR job.
But a full map stage is not a superstep, the whole computation is a done in
one mapping phase.
So a map task in MR corresponds to a computation phase in a superstep. Once
the c
Hi Praveen,
Answers inline. Hope that helps!
Avery
On 12/8/11 10:16 PM, Praveen Sripati wrote:
Hi,
I know about MapReduce/Hadoop and trying to get myself around
BSP/Hama-Giraph by comparing MR and BSP.
- Map Phase in MR is similar to Computation Phase in BSP. BSP allows
for process to ex
Hi Praveen,
Answers inline. Hope that helps!
Avery
On 12/8/11 10:16 PM, Praveen Sripati wrote:
Hi,
I know about MapReduce/Hadoop and trying to get myself around
BSP/Hama-Giraph by comparing MR and BSP.
- Map Phase in MR is similar to Computation Phase in BSP. BSP allows
for process to ex
11 matches
Mail list logo