Hello,

Just wanted to start the discussion about BSP on MPI and BSP for
shared memory architecture. GD already showed the 10x performance
improvement of Apache Hama applications using MPI + InfiniBand without
any changes.

Another is about BSP for shared memory architecture. As you know, some
algorithms generates lots of communications and all-reduce operations
among processors. To solve these problem, we might want to add new BSP
package for shared-memory.

What do you think?

-- 
Best Regards, Edward J. Yoon

Reply via email to