Point-to-Point Performance

So far, we conducted experiments on these different homogeneous protocols: Shmem, TCP over Gigabit-Ethernet, MX/GM/BIP over Myrinet and SISCI over SCI.

• SHARED MEMORY: Comparisons between MPICH-Madeleine, MPICH-SHMEM, MPICH-P4 w/Shmem, SCI-MPICH and MPICH-GM

  Latency	Bandwidth

• GigaBit Ethernet Network: Comparisons between MPICH-Madeleine and MPICH-P4

  Latency	Bandwidth

• SCI Network: Comparisons between MPICH-Madeleine and SCI-MPICH

  Latency	Bandwidth

• Myrinet Network: Comparisons between MPICH-Madeleine (both GM and MX) and MPICH-GM

  Latency	Bandwidth

HPL Performance

• 2-Nodes Configuration: Comparisons between MPICH-Madeleine, MPICH-P4, SCI-MPICH and MPICH-GM

  With SMT (4 processes/node)	Without SMT (2 processes/node)

• 4-Nodes Configuration: Comparisons between MPICH-Madeleine, MPICH-P4, and MPICH-GM

  With SMT (4 processes/node)