Best Point-to-Point Message-Passing Performance
on the Compaq AlphaServer SC
These results represent the highest measured bandwidths for each
of the experiments:
0 to 1 (intranode)
0 to 1, 2 to 3 simultaneously (intranode)
0 to 4 (internode)
0 to 4, 1 to 5, 2 to 6, 3 to 7 simultaneously (internode)
0 to 1 to 2 to 3 to 4 to 5 to 6 to 7 to 0 simultaneously (both inter and intra node comunication)
for the single iteration, nonoverlap communication protocols.
Results are presented for bidirectional and
unidirectional protocols separately, using MPI and SHMEM.
Note that the unidirectional bandwidth is also the "bidirectional"
bandwidth when using the optimal unidirectional protocol to complete
a swap. For some situations, this is a larger bandwidth (faster
protocol) than when using the optimal bidirectional protocol.
Patrick H. Worley / (
worleyph@ornl.gov)
