What a re s ome o f t he i mprovements i n g oing f rom t he
Harperstown chip to the Nehalem?
Th e n umber o f M PI me ssages t hat a n N IC c an ha ndle e ach
second is becoming more and more of a bottleneck on multicore
systems. Th e user can reduct the number of messages sent by
assuring t he b est de composition o n t he n ode. Giv en a n a llMPI app lication u sing 8 M PI t asks o n t he n ode, h ow w ould
the number of messages for one halo exchange change in the
following three scenarios?
a. Th e MPI tasks on the node are not neighbors of other MPI
tasks on the node.
b. Th e MPI tasks on the node are arranged in a 2 × 4 × 1 plane
of a grid.
c. Th e MPI tasks on the node are arranged in a 2 × 2 × 2 cube.
Already registered? Login
Not Account? Sign up
Enter your email address to reset your password
Back to Login? Click here