MapReduce instead of MPI
18 views
Skip to first unread message
Leo
Mar 10, 2009, 10:16:00 AM3/10/09
to PSVM
I didn't go through the paper and implemention in detail. MPI is used
for the data communication among cluster machines in PSVM here. Is it
possible to use MapReduce directly without MPI?
for the data communication among cluster machines in PSVM here. Is it
possible to use MapReduce directly without MPI?
Hongjie Bai
Mar 15, 2009, 11:38:14 PM3/15/09
to ps...@googlegroups.com
Hi Leo,
We haven't found a way to implement PSVM using mapreduce because it is not so flexible as MPI.
PSVM requires communication several times in an iteration while there is only one chance
to communicate in mapreduce.
We haven't found a way to implement PSVM using mapreduce because it is not so flexible as MPI.
PSVM requires communication several times in an iteration while there is only one chance
to communicate in mapreduce.
Edward J. Yoon
Apr 10, 2009, 2:19:51 AM4/10/09
to PSVM
Why not use bigtable for the communications between nodes?
On Mar 16, 12:38 pm, Hongjie Bai <baihong...@gmail.com> wrote:
> Hi Leo,
>
> We haven't found a way to implement PSVM using mapreduce because it is not
> so flexible as MPI.
> PSVM requires communication several times in an iteration while there is
> only one chance
> to communicate in mapreduce.
>
On Mar 16, 12:38 pm, Hongjie Bai <baihong...@gmail.com> wrote:
> Hi Leo,
>
> We haven't found a way to implement PSVM using mapreduce because it is not
> so flexible as MPI.
> PSVM requires communication several times in an iteration while there is
> only one chance
> to communicate in mapreduce.
>
Reply all
Reply to author
Forward
0 new messages